😏

profiles/README.md

@@ -0,0 +1,33 @@
+# Bundled profiles
+
+The 8.0.0 portfolio contains eight featured workload profiles and four explicitly experimental labs. Every MMO-owned document in this development generation uses schema `8`; profiles also require `version = "8.0.0"`. Earlier generations, retired profile IDs, and old catalog keys are invalid, with no compatibility reader, alias, migration, fallback host, or automatic rewrite. External formats owned by Codex app-server, MCP, providers, and Switchyard keep their upstream identifiers. Spawn admission is based only on simultaneous active capacity: terminal, cold-paused, and suspended workers release their role, route, resource, child, and writer slots for later or resumed work, and continuation performs fresh admission.
+
+Every root and Agent-MCP role is hosted by a Codex 0.149.0 app-server on a private Unix socket while executing. Goal roles use token budgets with compiled ceilings; turn roles have no task wall clock. Stall intervals are warnings, not interrupts, and the model is never told to measure time. Per-target `controls` grants expose only the admitted inspect/trace/steer/interrupt/pause/continue/detach/stop/finalize/compact/respond/effort/fork actions. Cold pause retires MCP hosts while retaining exact threads and evidence; detach leaves them live. Durable events, terminal history, partial results, and isolated patches survive client/controller/provider loss; they do not weaken the role's trust, permission, or verification policy.
+
+## Featured
+
+| Profile | Practical purpose | Composition | Engine |
+|---|---|---|---|
+| `adaptive-engineering` | General engineering with selective reconnaissance, isolated implementation, and independent review | Sol root; Luna, DeepSeek V4 Pro, Opus 5 | Hybrid |
+| `codex-harness-team` | ChatGPT-only control showing the value of context isolation and role specialization | Sol root; Luna, Terra, fresh Sol critic | Hybrid |
+| `high-confidence-debugging` | Reproduce first, repair second, then attempt to falsify the fix | DeepSeek V4 Pro root; Sonnet 5, Sol | Agent MCP |
+| `research-backed-engineering` | Current-source implementation with claim-level citations and source-to-code audit | MiniMax M3 root; Luna, Sonnet 5 | Agent MCP |
+| `secure-change` | Threat framing, independent vulnerability hunting, empirical reproduction, and remediation | Opus 5 root; GLM-5.3 via Z.AI Coding Plan, DeepSeek V4 Pro | Agent MCP |
+| `visual-engineering` | Reference-to-code UI work with real browser rendering and fresh multimodal verification | Sol root; Terra, fresh Sol verifier | Hybrid |
+| `contract-first-refactoring` | Freeze behavior, derive independent contract tests, refactor, and verify compatibility | Sonnet 5 root; MiniMax M3, Luna, DeepSeek V4 Pro | Agent MCP |
+| `incident-hypothesis-triage` | Read-only incident analysis that separates evidence collection from causal challenge | GLM-5.3 Coding Plan root; DeepSeek V4 Flash, pinned Nemotron | Agent MCP |
+
+## Labs
+
+| Profile | Hypothesis under test | Composition | Engine |
+|---|---|---|---|
+| `access-efficient-escalation-lab` | Cheap/local bounded work plus deliberate escalation can reduce paid usage without losing quality | Luna root; local Qwen3.5-9B, DeepSeek V4 Flash, GLM-5.3 | Agent MCP |
+| `route-resilience-lab` | The same maker/model through independent access routes can expose route drift and outage behavior | Terra root; DeepSeek V4 Pro through Go and pinned OpenRouter | Agent MCP |
+| `bounded-research-organization-lab` | A strictly bounded research hierarchy can outperform flat fan-out on long corpora | Sol root; Kimi K3 lead, MiniMax M3 scouts | Agent MCP |
+| `competing-implementations-lab` | Two isolated implementations selected by an objective contract can beat a single implementation | Terra judge; DeepSeek V4 Pro and Sonnet 5 candidates | Agent MCP |
+
+Labs are packaged experiments, not recommended defaults. Their README and evaluation suite state the hypothesis, matched controls, and promotion gate.
+
+Every pack contains only static TOML, Markdown instructions, and JSON result contracts. It does not contain credentials, endpoint definitions, Tool MCP server definitions, executable workflow code, or compatibility shims. Exact route-qualified model keys make the model maker, access route, and wire protocol inspectable through the catalog.
+
+See [the full profile guide](../docs/PROFILES.md) for role graphs, availability behavior, trust boundaries, and selection guidance.

profiles/access-efficient-escalation-lab/README.md

@@ -0,0 +1,7 @@
+# Access-Efficient Escalation Lab
+
+This lab tests whether cheap or local work can reduce scarce-tier use without unacceptable quality loss. Luna remains the responsible root; local Qwen receives only runtime-generated literal operations; DeepSeek Flash handles one objective isolated patch; GLM is a deliberate read-only escalation.
+
+The local route is optional. Weak-model tasks have no free-form prompt surface, no network, no writes, no spawning, a single GPU slot, and hash/line verification. The lab is not promoted by lower nominal cost alone: matched trials must show at least 50% lower API-equivalent or scarce-tier consumption with no more than two points of success loss.
+
+All three workers use persistent supervised Unix app-server threads. The root can inspect, trace, steer, interrupt, pause, continue, detach, finalize, fully stop, or cancel each one. Goal token ceilings or clock-free turns replace per-spawn model timeouts; stall warnings never interrupt, and suspension retains terminal history, trace, partial evidence, and any isolated patch. The Qwen role remains low-trust and cannot control or spawn anything.

profiles/access-efficient-escalation-lab/agents/access-router.md

@@ -0,0 +1 @@
+Own the critical path and canonical workspace. Use the local scout only for one structured literal operation, never free-form reasoning. Use DeepSeek Flash for one bounded routine patch with objective tests. Escalate to GLM only when recorded evidence shows unresolved ambiguity, security risk, or failed routine verification. Continue useful root work while independent tasks run, explicitly accept or reject results, and integrate isolated patches only after tests. Record which scarce-tier calls were avoided; this lab makes no quality claim until matched trials satisfy its gate.

profiles/access-efficient-escalation-lab/agents/flagship-escalation.md

@@ -0,0 +1 @@
+Address only the explicitly documented escalation trigger. Analyze the unresolved high-impact question read-only, cite repository evidence, challenge the routine conclusion, and return a decision recommendation with falsification tests. Do not edit or manufacture certainty.

profiles/access-efficient-escalation-lab/agents/literal-scout.md

@@ -0,0 +1 @@
+The runtime supplies one literal operation. Perform only that operation and return exact path, hash, line-range, and excerpt evidence or a hash-linked summary of supplied text. Do not infer intent, diagnose, design, recommend, execute arbitrary commands, edit files, browse, or spawn.

profiles/access-efficient-escalation-lab/agents/routine-engineer.md

@@ -0,0 +1 @@
+Implement only the bounded routine change in the assigned isolated scope. Run objective tests and return correlated command evidence. Do not expand the task or assume the root will integrate the patch.

profiles/access-efficient-escalation-lab/contracts/engineering.json

@@ -0,0 +1,13 @@
+{
+  "title": "Economical routine engineering patch",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "summary": {"type": "string", "minLength": 1},
+    "files": {"type": "array", "items": {"type": "string"}, "maxItems": 30, "uniqueItems": true},
+    "commands": {"type": "array", "maxItems": 30, "items": {"type": "object", "additionalProperties": false, "properties": {"command": {"type": "string", "minLength": 1}, "exit_code": {"type": "integer"}, "observation": {"type": "string", "minLength": 1}}, "required": ["command", "exit_code", "observation"]}},
+    "risks": {"type": "array", "items": {"type": "string"}, "maxItems": 20},
+    "blockers": {"type": "array", "items": {"type": "string"}, "maxItems": 20}
+  },
+  "required": ["summary", "files", "commands", "risks", "blockers"]
+}

profiles/access-efficient-escalation-lab/contracts/escalation.json

@@ -0,0 +1,13 @@
+{
+  "title": "Evidence-triggered difficult-reasoning escalation",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "trigger": {"type": "string", "minLength": 1},
+    "recommendation": {"type": "string", "minLength": 1},
+    "evidence": {"type": "array", "items": {"type": "string"}, "minItems": 1, "maxItems": 50},
+    "falsification_tests": {"type": "array", "items": {"type": "string"}, "minItems": 1, "maxItems": 30},
+    "uncertainties": {"type": "array", "items": {"type": "string"}, "maxItems": 30}
+  },
+  "required": ["trigger", "recommendation", "evidence", "falsification_tests", "uncertainties"]
+}

profiles/access-efficient-escalation-lab/contracts/literal-evidence.json

@@ -0,0 +1,31 @@
+{
+  "title": "Mechanically checked literal evidence",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "operation": {"type": "string", "enum": ["locate", "references", "extract", "summarize_supplied"]},
+    "input_sha256": {"type": ["string", "null"], "pattern": "^[0-9a-f]{64}$"},
+    "evidence": {
+      "type": "array",
+      "maxItems": 200,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "path": {"type": "string", "minLength": 1},
+          "sha256": {"type": "string", "pattern": "^[0-9a-f]{64}$"},
+          "start_line": {"type": "integer", "minimum": 1},
+          "end_line": {"type": "integer", "minimum": 1},
+          "excerpt": {"type": "string"}
+        },
+        "required": ["path", "sha256", "start_line", "end_line", "excerpt"]
+      }
+    },
+    "summary_points": {"type": "array", "items": {"type": "string"}, "maxItems": 50}
+  },
+  "required": ["operation", "input_sha256", "evidence", "summary_points"],
+  "allOf": [
+    {"if": {"properties": {"operation": {"enum": ["locate", "references", "extract"]}}, "required": ["operation"]}, "then": {"properties": {"input_sha256": {"const": null}, "evidence": {"minItems": 1}, "summary_points": {"maxItems": 0}}}},
+    {"if": {"properties": {"operation": {"const": "summarize_supplied"}}, "required": ["operation"]}, "then": {"properties": {"input_sha256": {"type": "string", "pattern": "^[0-9a-f]{64}$"}, "evidence": {"maxItems": 0}, "summary_points": {"minItems": 1}}}}
+  ]
+}

profiles/access-efficient-escalation-lab/profile.toml

@@ -0,0 +1,111 @@
+schema_version = 8
+id = "access-efficient-escalation-lab"
+version = "8.0.0"
+display_name = "Access-Efficient Escalation Lab"
+description = "An experimental Luna root routes literal reconnaissance to a contained local model, routine patches to economical DeepSeek Flash, and difficult read-only analysis to GLM only when justified."
+tags = ["cost", "local-first", "escalation", "lab"]
+maturity = "lab"
+root = "access_router"
+smoke = "smoke.toml"
+
+[agents.access_router]
+kind = "root"
+description = "Productive routing root, canonical writer, and escalation authority"
+model = "codex_chatgpt_builtin__gpt_5_6_luna"
+reasoning = "medium"
+allowed_reasoning_efforts = ["low", "medium", "high"]
+execution_mode = "goal"
+goal_token_budget = 250000
+max_goal_token_budget = 400000
+stall_warning_seconds = 3600
+finalization_grace_seconds = 1200
+plan_reasoning = "medium"
+permissions = "workspace-write"
+can_spawn = ["literal_scout", "routine_engineer", "flagship_escalation"]
+max_active = 1
+trust = "high"
+verification = "always"
+instructions = "agents/access-router.md"
+
+[agents.literal_scout]
+description = "Mechanically contained local model for literal repository evidence only"
+model = "llama_cpp_local_openai_chat__qwen3_5_9b"
+reasoning = "none"
+allowed_reasoning_efforts = ["none"]
+execution_mode = "turn"
+stall_warning_seconds = 1800
+finalization_grace_seconds = 600
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+max_task_chars = 2500
+allowed_task_kinds = ["locate", "references", "extract", "summarize_supplied"]
+trust = "low"
+verification = "always"
+network_access = false
+web_search = "disabled"
+contract_enforcement = "strict"
+resource_group = "local_gpu_0"
+output_contract = "contracts/literal-evidence.json"
+instructions = "agents/literal-scout.md"
+
+[agents.routine_engineer]
+description = "Economical isolated worker for one bounded routine patch"
+model = "opencode_go_openai_chat__deepseek_v4_flash"
+reasoning = "high"
+allowed_reasoning_efforts = ["low", "high", "max"]
+execution_mode = "turn"
+stall_warning_seconds = 1800
+finalization_grace_seconds = 1200
+permissions = "workspace-write"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["implement", "debug", "test", "refactor"]
+trust = "high"
+verification = "always"
+contract_enforcement = "strict"
+output_contract = "contracts/engineering.json"
+instructions = "agents/routine-engineer.md"
+
+[agents.flagship_escalation]
+description = "Read-only difficult-reasoning escalation used only after explicit trigger evidence"
+model = "zai_coding_responses__glm_5_3"
+reasoning = "max"
+allowed_reasoning_efforts = ["high", "max"]
+execution_mode = "turn"
+stall_warning_seconds = 1800
+finalization_grace_seconds = 900
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["analysis", "review", "security_review"]
+trust = "high"
+verification = "root_adjudication"
+contract_enforcement = "strict"
+output_contract = "contracts/escalation.json"
+instructions = "agents/flagship-escalation.md"
+
+[agents.access_router.controls.literal_scout]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.access_router.controls.routine_engineer]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.access_router.controls.flagship_escalation]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[coordination]
+mode = "access_efficient_escalation_experiment"
+orchestration = "mcp"
+max_active_agents = 3
+max_depth = 1
+max_children_per_agent = 3
+max_active_writers = 1
+reject_ancestor_role = true
+wait_policy = "dependency_only"
+write_conflict_policy = "reject"
+contradiction_policy = "primary_evidence"
+result_visibility = "ancestors"

profiles/access-efficient-escalation-lab/smoke.toml

@@ -0,0 +1,36 @@
+schema_version = 8
+
+[[tasks]]
+agent = "access_router"
+task_kind = "analysis"
+task = "Inspect this lab read-only and state the mechanical escalation and weak-model containment rules."
+mode = "read-only"
+wall_timeout_seconds = 600
+
+[[tasks]]
+agent = "literal_scout"
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 600
+
+[tasks.literal_task]
+operation = "locate"
+needle = "MMO_SCHEMA_VERSION"
+paths = ["libexec/mmo_profiles.py"]
+max_results = 10
+
+[[tasks]]
+agent = "routine_engineer"
+task_kind = "test"
+task = "Inspect one existing test read-only, run one focused validation command, and return the engineering JSON with no file changes."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900
+
+[[tasks]]
+agent = "flagship_escalation"
+task_kind = "analysis"
+task = "Treat uncertainty about whether schema v1 remains accepted as the explicit escalation trigger. Analyze it read-only and return only the escalation JSON with exact evidence and a falsification test."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900

profiles/adaptive-engineering/README.md

@@ -0,0 +1,16 @@
+# Adaptive Engineering
+
+Use this flagship profile for difficult general software engineering where selective reconnaissance, a separable implementation, or independent review can improve a strong root. Do not use it for trivial edits or as an automatic four-agent fan-out.
+
+```text
+Sol orchestrator (active root)
+├── Luna repo scout (native, read-only)
+├── DeepSeek V4 Pro implementation specialist (MCP, one leased write scope)
+└── Claude Opus 5 adversarial reviewer (MCP, read-only)
+```
+
+The root decides dynamically which roles justify their cost and continues critical-path work while they run. MCP jobs mechanically receive exact models, goal/turn lifecycle, warning-only stalls, scopes, and strict contracts. The native scout's result format and role boundary are advisory. Root avoidance of an active worker scope is an instruction because the root is not covered by MCP worker leases.
+
+The implementation specialist and adversarial reviewer run on persistent supervised app-server threads; the root controls both, and the two roles may challenge, steer, pause, or return each other to focused work through exact reciprocal control edges. Transport loss suspends instead of erasing them. The Luna scout remains a Codex-native participant, so MMO does not claim app-server lifecycle or live-control enforcement for that branch.
+
+Expected cost is medium to high and latency is medium. Optional worker failure reduces coverage but need not block a safe root-owned result. There is no automatic provider substitution.

profiles/adaptive-engineering/agents/adversarial-reviewer.md

@@ -0,0 +1 @@
+Act as one strong independent peer, not a ceremonial council. For an architecture consultation, identify competing feasible approaches, hidden assumptions, falsification tests, and the evidence that should decide. For a post-change review, attack correctness, concurrency, security, compatibility, and test gaps. Ground findings in exact evidence and distinguish confirmed defects from hypotheses. Do not decide by model reputation or majority vote.

profiles/adaptive-engineering/agents/implementation-specialist.md

@@ -0,0 +1 @@
+Implement only the bounded branch and write scope assigned by the root. Work from stated invariants, run focused tests, and return a minimal reviewable patch with exact command evidence. Do not broaden the design, touch unrelated files, or assume that your isolated worktree will be merged. State unresolved risks and blockers plainly; the root alone accepts and integrates the patch.

profiles/adaptive-engineering/agents/orchestrator.md

@@ -0,0 +1,3 @@
+Own the critical path and canonical workspace. Decompose only independent work whose expected value exceeds its handoff cost. Launch the Luna scout early for bounded repository facts, use the DeepSeek specialist for at most one disjoint and objectively testable patch, and use the Opus peer either before implementation for a material architecture fork or after the diff for adversarial review. Continue diagnosing or implementing non-overlapping work while workers run; wait only at a genuine dependency barrier.
+
+Treat worker output as claims. Read each result, inspect evidence and artifacts, then explicitly accept or reject it. For a writable worker, review the binary patch and tests before explicit integration. Never equate reading with acceptance. Resolve disagreements through repository evidence, commands, tests, or specifications—not voting or model reputation. If an optional route is unavailable, record the degraded topology and complete the task yourself when safe.

profiles/adaptive-engineering/agents/repo-scout.md

@@ -0,0 +1 @@
+Perform only the bounded reconnaissance requested. Report exact paths, symbols, line locations, tests, commands, and uncertainties. Do not design the solution, edit files, or duplicate the root's critical-path investigation. Make every consequential claim easy for the root to verify from repository evidence.

profiles/adaptive-engineering/contracts/engineering.json

@@ -0,0 +1,27 @@
+{
+  "title": "Bounded isolated engineering patch",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "summary": {"type": "string", "minLength": 1},
+    "changes": {"type": "array", "items": {"type": "string"}, "maxItems": 50},
+    "files": {"type": "array", "items": {"type": "string"}, "maxItems": 50, "uniqueItems": true},
+    "commands": {
+      "type": "array",
+      "maxItems": 30,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "command": {"type": "string", "minLength": 1},
+          "exit_code": {"type": "integer"},
+          "observation": {"type": "string", "minLength": 1}
+        },
+        "required": ["command", "exit_code", "observation"]
+      }
+    },
+    "risks": {"type": "array", "items": {"type": "string"}, "maxItems": 30},
+    "blockers": {"type": "array", "items": {"type": "string"}, "maxItems": 30}
+  },
+  "required": ["summary", "changes", "files", "commands", "risks", "blockers"]
+}

profiles/adaptive-engineering/contracts/evidence.json

@@ -0,0 +1,26 @@
+{
+  "additionalProperties": false,
+  "properties": {
+    "status": {"enum": ["ok", "conflict", "blocked"], "type": "string"},
+    "evidence": {
+      "items": {
+        "additionalProperties": false,
+        "properties": {
+          "claim": {"minLength": 1, "type": "string"},
+          "source": {"minLength": 1, "type": "string"},
+          "location": {"type": "string"},
+          "observation": {"minLength": 1, "type": "string"}
+        },
+        "required": ["claim", "source", "observation"],
+        "type": "object"
+      },
+      "maxItems": 100,
+      "type": "array"
+    },
+    "conflicts": {"items": {"type": "string"}, "maxItems": 30, "type": "array"},
+    "limitations": {"items": {"type": "string"}, "maxItems": 30, "type": "array"}
+  },
+  "required": ["status", "evidence", "conflicts", "limitations"],
+  "title": "Repository evidence report",
+  "type": "object"
+}

profiles/adaptive-engineering/contracts/review.json

@@ -0,0 +1,38 @@
+{
+  "title": "Independent architecture or diff review",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "verdict": {"type": "string", "enum": ["pass", "concerns", "fail", "blocked"]},
+    "review_mode": {"type": "string", "enum": ["architecture", "post_change"]},
+    "findings": {
+      "type": "array",
+      "maxItems": 50,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "severity": {"type": "string", "enum": ["critical", "high", "medium", "low", "note"]},
+          "claim": {"type": "string", "minLength": 1},
+          "evidence": {"type": "string", "minLength": 1},
+          "location": {"type": "string"},
+          "falsification": {"type": "string", "minLength": 1}
+        },
+        "required": ["severity", "claim", "evidence", "falsification"]
+      }
+    },
+    "alternatives": {"type": "array", "items": {"type": "string"}, "maxItems": 20},
+    "blockers": {"type": "array", "items": {"type": "string"}, "maxItems": 20}
+  },
+  "required": ["verdict", "review_mode", "findings", "alternatives", "blockers"],
+  "allOf": [
+    {
+      "if": {"properties": {"verdict": {"const": "pass"}}, "required": ["verdict"]},
+      "then": {"properties": {"findings": {"maxItems": 0}, "blockers": {"maxItems": 0}}}
+    },
+    {
+      "if": {"properties": {"verdict": {"const": "blocked"}}, "required": ["verdict"]},
+      "then": {"properties": {"blockers": {"minItems": 1}}}
+    }
+  ]
+}

profiles/adaptive-engineering/profile.toml

@@ -0,0 +1,116 @@
+schema_version = 8
+id = "adaptive-engineering"
+version = "8.0.0"
+display_name = "Adaptive Engineering"
+description = "A productive Sol root selectively combines fast native reconnaissance, an economical isolated implementation branch, and one strong cross-provider peer."
+tags = ["engineering", "adaptive", "hybrid", "featured"]
+maturity = "featured"
+root = "orchestrator"
+smoke = "smoke.toml"
+
+[agents.orchestrator]
+kind = "root"
+description = "Critical-path engineer, canonical writer, evidence adjudicator, and integration authority"
+model = "codex_chatgpt_builtin__gpt_5_6_sol"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh", "max"]
+execution_mode = "goal"
+goal_token_budget = 600000
+max_goal_token_budget = 1000000
+stall_warning_seconds = 3600
+finalization_grace_seconds = 1800
+plan_reasoning = "high"
+permissions = "workspace-write"
+can_spawn = ["repo_scout", "implementation_specialist", "adversarial_reviewer"]
+max_active = 1
+trust = "high"
+verification = "risk_based"
+attachments_allowed = true
+requires_modalities = ["text", "image"]
+instructions = "agents/orchestrator.md"
+
+[agents.repo_scout]
+description = "Fast native read-only repository and test reconnaissance"
+model = "codex_chatgpt_builtin__gpt_5_6_luna"
+reasoning = "medium"
+allowed_reasoning_efforts = ["low", "medium", "high"]
+permissions = "read-only"
+backends = ["native"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["analysis", "locate", "references", "test_analysis"]
+trust = "normal"
+verification = "material_changes"
+contract_enforcement = "warn"
+output_contract = "contracts/evidence.json"
+instructions = "agents/repo-scout.md"
+
+[agents.implementation_specialist]
+description = "One bounded, disjoint, objectively testable implementation branch"
+model = "opencode_go_openai_chat__deepseek_v4_pro"
+reasoning = "high"
+allowed_reasoning_efforts = ["high", "max"]
+execution_mode = "goal"
+goal_token_budget = 250000
+max_goal_token_budget = 400000
+stall_warning_seconds = 1800
+finalization_grace_seconds = 1200
+permissions = "workspace-write"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["analysis", "implement", "debug", "test", "refactor"]
+trust = "high"
+verification = "always"
+contract_enforcement = "strict"
+output_contract = "contracts/engineering.json"
+instructions = "agents/implementation-specialist.md"
+
+[agents.adversarial_reviewer]
+description = "One architecture consultation or post-change adversarial review"
+model = "opencode_zen_anthropic_messages__claude_opus_5"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh", "max"]
+execution_mode = "turn"
+stall_warning_seconds = 1800
+finalization_grace_seconds = 900
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["architecture", "review", "verification", "security_review"]
+trust = "adversarial_reviewer"
+verification = "root_adjudication"
+contract_enforcement = "strict"
+output_contract = "contracts/review.json"
+instructions = "agents/adversarial-reviewer.md"
+
+[agents.orchestrator.controls.repo_scout]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.orchestrator.controls.implementation_specialist]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.orchestrator.controls.adversarial_reviewer]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.implementation_specialist.controls.adversarial_reviewer]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.adversarial_reviewer.controls.implementation_specialist]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[coordination]
+mode = "adaptive_engineering"
+orchestration = "hybrid"
+max_active_agents = 4
+max_depth = 1
+max_children_per_agent = 3
+max_active_writers = 1
+native_max_concurrent_threads = 1
+native_nested_delegation = false
+reject_ancestor_role = true
+wait_policy = "dependency_only"
+write_conflict_policy = "reject"
+contradiction_policy = "primary_evidence"
+result_visibility = "ancestors"

profiles/adaptive-engineering/smoke.toml

@@ -0,0 +1,32 @@
+schema_version = 8
+
+[[tasks]]
+agent = "orchestrator"
+task_kind = "analysis"
+task = "Inspect this repository read-only and explain when this profile should select each worker and when the root should continue alone."
+mode = "read-only"
+wall_timeout_seconds = 600
+
+[[tasks]]
+agent = "repo_scout"
+task_kind = "locate"
+task = "Locate the profile compiler and its tests. Return exact paths and evidence in the required JSON shape; do not propose changes."
+backend = "native"
+mode = "read-only"
+wall_timeout_seconds = 600
+
+[[tasks]]
+agent = "implementation_specialist"
+task_kind = "analysis"
+task = "Inspect the profile schema validation read-only. Return the required engineering JSON with an empty changes/files list and any focused validation command evidence."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900
+
+[[tasks]]
+agent = "adversarial_reviewer"
+task_kind = "review"
+task = "Review the adaptive profile definition for one concrete orchestration risk. Return only the required review JSON grounded in repository evidence."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900

profiles/bounded-research-organization-lab/README.md

@@ -0,0 +1,7 @@
+# Bounded Research Organization Lab
+
+This lab compares two ways to research the same large corpus: a flat Sol-to-two-scout fan-out and a depth-two Kimi-to-two-scout hierarchy. MiniMax scouts, prompts, source questions, and trial tasks remain matched so the handoff itself can be evaluated.
+
+Kimi may spawn at most two scouts and the whole session permits only three worker launches. The root continues repository work instead of waiting. Promotion requires evidence that synthesis quality or root productivity outweighs extra latency, duplicate work, and information lost across the additional handoff.
+
+Kimi and every MiniMax scout use persistent supervised Unix app-server threads with goal/turn lifecycle, warning-only stalls, terminal history, and retained partial evidence. The root can control either layer, while Kimi can control only its scouts; detach, pause, interruption, continuation, finalization, full stop, and cancellation preserve the bounded hierarchy instead of replacing it with polling or model-authored clock checkpoints.

profiles/bounded-research-organization-lab/agents/engineering-root.md

@@ -0,0 +1 @@
+Remain productive on repository work while research runs. For a hierarchical trial, spawn Kimi once and let it allocate at most two bounded scouts; for a flat control, spawn the same scouts directly with identical questions. Do not mix topologies within a matched trial. Re-fetch pivotal claims, explicitly accept or reject dossiers, and measure duplicate work, handoff loss, latency, and source quality before claiming hierarchy helped.

profiles/bounded-research-organization-lab/agents/research-lead.md

@@ -0,0 +1 @@
+Organize a large-source research question into at most two independent bounded scout tasks. Launch them together when useful and continue synthesizing supplied material while they run. Read and explicitly accept or reject their results. Return a source-linked synthesis that identifies handoff loss, conflicts, and claims the engineering root must re-fetch. Do not edit the repository or exceed the bounded graph.

profiles/bounded-research-organization-lab/agents/source-scout.md

@@ -0,0 +1 @@
+Research one assigned question using current primary sources. Return dated URLs, retrieval timestamps, direct evidence, conflicts, and limitations. Do not broaden scope, edit files, or synthesize the parent question.

profiles/bounded-research-organization-lab/contracts/research-synthesis.json

@@ -0,0 +1,14 @@
+{
+  "title": "Bounded hierarchical research synthesis",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "accepted_claims": {"type": "array", "items": {"type": "string"}, "maxItems": 100},
+    "rejected_claims": {"type": "array", "items": {"type": "string"}, "maxItems": 100},
+    "source_urls": {"type": "array", "items": {"type": "string", "format": "uri"}, "maxItems": 100, "uniqueItems": true},
+    "handoff_loss": {"type": "array", "items": {"type": "string"}, "maxItems": 30},
+    "duplicate_work": {"type": "array", "items": {"type": "string"}, "maxItems": 30},
+    "root_refetch_required": {"type": "array", "items": {"type": "string"}, "maxItems": 30}
+  },
+  "required": ["accepted_claims", "rejected_claims", "source_urls", "handoff_loss", "duplicate_work", "root_refetch_required"]
+}

profiles/bounded-research-organization-lab/contracts/source-dossier.json

@@ -0,0 +1,12 @@
+{
+  "title": "Bounded scout source dossier",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "question": {"type": "string", "minLength": 1},
+    "claims": {"type": "array", "maxItems": 50, "items": {"type": "object", "additionalProperties": false, "properties": {"claim": {"type": "string", "minLength": 1}, "source_url": {"type": "string", "format": "uri"}, "source_date": {"type": "string", "format": "date"}, "retrieved_at": {"type": "string", "format": "date-time"}, "evidence": {"type": "string", "minLength": 1}}, "required": ["claim", "source_url", "source_date", "retrieved_at", "evidence"]}},
+    "conflicts": {"type": "array", "items": {"type": "string"}, "maxItems": 30},
+    "limitations": {"type": "array", "items": {"type": "string"}, "maxItems": 30}
+  },
+  "required": ["question", "claims", "conflicts", "limitations"]
+}

profiles/bounded-research-organization-lab/profile.toml

@@ -0,0 +1,95 @@
+schema_version = 8
+id = "bounded-research-organization-lab"
+version = "8.0.0"
+display_name = "Bounded Research Organization Lab"
+description = "An experimental Sol root compares flat research fan-out with a strictly bounded Kimi-led hierarchy over long source corpora."
+tags = ["research", "nested-delegation", "long-context", "lab"]
+maturity = "lab"
+root = "engineering_root"
+smoke = "smoke.toml"
+
+[agents.engineering_root]
+kind = "root"
+description = "Productive repository engineer and final evidence adjudicator"
+model = "codex_chatgpt_builtin__gpt_5_6_sol"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh", "max"]
+execution_mode = "goal"
+goal_token_budget = 500000
+max_goal_token_budget = 800000
+stall_warning_seconds = 3600
+finalization_grace_seconds = 1800
+plan_reasoning = "high"
+permissions = "workspace-write"
+can_spawn = ["research_lead", "source_scout"]
+max_active = 1
+trust = "high"
+verification = "always"
+instructions = "agents/engineering-root.md"
+
+[agents.research_lead]
+description = "Long-context Kimi research organizer with exactly two possible child scouts"
+model = "opencode_go_openai_chat__kimi_k3"
+reasoning = "max"
+allowed_reasoning_efforts = ["max"]
+execution_mode = "goal"
+goal_token_budget = 300000
+max_goal_token_budget = 500000
+stall_warning_seconds = 1800
+finalization_grace_seconds = 1800
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = ["source_scout"]
+max_active = 1
+max_children = 2
+allowed_task_kinds = ["research_organization", "research_synthesis"]
+trust = "high"
+verification = "root_adjudication"
+network_access = false
+web_search = "live"
+contract_enforcement = "strict"
+output_contract = "contracts/research-synthesis.json"
+instructions = "agents/research-lead.md"
+
+[agents.source_scout]
+description = "Bounded MiniMax source scout usable in either flat or hierarchical trials"
+model = "opencode_zen_openai_chat__minimax_m3"
+reasoning = "none"
+allowed_reasoning_efforts = ["none"]
+execution_mode = "turn"
+stall_warning_seconds = 1800
+finalization_grace_seconds = 900
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = []
+max_active = 2
+allowed_task_kinds = ["research", "source_verification"]
+trust = "high"
+verification = "always"
+network_access = false
+web_search = "live"
+contract_enforcement = "strict"
+output_contract = "contracts/source-dossier.json"
+instructions = "agents/source-scout.md"
+
+[agents.engineering_root.controls.research_lead]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.engineering_root.controls.source_scout]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.research_lead.controls.source_scout]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[coordination]
+mode = "bounded_hierarchical_research_experiment"
+orchestration = "mcp"
+max_active_agents = 4
+max_depth = 2
+max_children_per_agent = 2
+max_active_writers = 0
+reject_ancestor_role = true
+wait_policy = "dependency_only"
+write_conflict_policy = "reject"
+contradiction_policy = "primary_evidence"
+result_visibility = "ancestors"

profiles/bounded-research-organization-lab/smoke.toml

@@ -0,0 +1,24 @@
+schema_version = 8
+
+[[tasks]]
+agent = "engineering_root"
+task_kind = "analysis"
+task = "Inspect this lab read-only and contrast the matched flat and bounded hierarchical trial topologies."
+mode = "read-only"
+wall_timeout_seconds = 600
+
+[[tasks]]
+agent = "research_lead"
+task_kind = "research_synthesis"
+task = "Without spawning for this smoke task, synthesize the supplied profile README into the required research-synthesis JSON and identify what the root must independently verify."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900
+
+[[tasks]]
+agent = "source_scout"
+task_kind = "research"
+task = "Verify one current Python documentation claim from a primary source and return only the dated source-dossier JSON."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900

profiles/codex-harness-team/README.md

@@ -0,0 +1,7 @@
+# Codex Harness Team
+
+Use this featured profile for general engineering when ChatGPT-backed Codex access is the only required service. Sol owns the critical path and all writes; Luna scouts and Terra designs invariants in isolated native contexts; a fresh Sol context performs one mechanically supervised post-diff review.
+
+The profile demonstrates that context isolation, parallel read work, role specialization, and adversarial review can add value without changing model makers. Native controls remain advisory, so both native workers are read-only. If the fresh MCP critic cannot run, the root continues with visibly reduced independent verification.
+
+The fresh critic is the one supervised persistent app-server worker: the integrator can inspect, steer, interrupt, continue, finalize, fork, or cancel it, and its trace and partial evidence survive host loss. Luna and Terra remain native Codex participants; they intentionally demonstrate native context isolation and are not presented as mechanically app-server-controlled workers.

profiles/codex-harness-team/agents/fresh-critic.md

@@ -0,0 +1 @@
+Review the integrated diff from a fresh context. Attack correctness, hidden invariants, regressions, security, error handling, and test adequacy. Ground every finding in an exact location or captured command. Return pass only when no material finding remains; valid JSON alone is not evidence.

profiles/codex-harness-team/agents/integrator.md

@@ -0,0 +1 @@
+Own the critical path and canonical writes. Launch the fast repository scout and invariant designer together only when their questions are independent, then immediately continue diagnosis or implementation. Native participants are advisory read-only contexts: inspect their evidence rather than assuming MMO can supervise every native action. After the integrated diff is stable, use the fresh MCP Sol critic once. Read and explicitly accept or reject that result. Resolve conflicts with repository evidence and tests, and wait only at real barriers.

profiles/codex-harness-team/agents/invariant-designer.md

@@ -0,0 +1 @@
+Independently derive explicit invariants, failure cases, and discriminating tests from the task and repository. Do not edit files or prescribe implementation details unless required to make a test precise. Label assumptions and show how each proposed test would falsify a plausible wrong implementation.

profiles/codex-harness-team/agents/repo-scout.md

@@ -0,0 +1 @@
+Locate exact files, symbols, tests, and dependencies for the bounded question. Be fast and factual. Do not edit, design the full solution, or duplicate the root's implementation. Return evidence the root can inspect directly.

profiles/codex-harness-team/contracts/evidence.json

@@ -0,0 +1,26 @@
+{
+  "title": "Native context evidence",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "scope": {"type": "string", "minLength": 1},
+    "evidence": {
+      "type": "array",
+      "maxItems": 50,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "claim": {"type": "string", "minLength": 1},
+          "location": {"type": "string", "minLength": 1},
+          "observation": {"type": "string", "minLength": 1}
+        },
+        "required": ["claim", "location", "observation"]
+      }
+    },
+    "invariants": {"type": "array", "items": {"type": "string"}, "maxItems": 30},
+    "falsifying_tests": {"type": "array", "items": {"type": "string"}, "maxItems": 30},
+    "uncertainties": {"type": "array", "items": {"type": "string"}, "maxItems": 30}
+  },
+  "required": ["scope", "evidence", "invariants", "falsifying_tests", "uncertainties"]
+}

profiles/codex-harness-team/contracts/review.json

@@ -0,0 +1,44 @@
+{
+  "title": "Fresh-context integrated diff review",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "verdict": {"type": "string", "enum": ["pass", "concerns", "fail", "blocked"]},
+    "findings": {
+      "type": "array",
+      "maxItems": 50,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "severity": {"type": "string", "enum": ["critical", "high", "medium", "low", "note"]},
+          "location": {"type": "string", "minLength": 1},
+          "claim": {"type": "string", "minLength": 1},
+          "evidence": {"type": "string", "minLength": 1},
+          "correction": {"type": "string", "minLength": 1}
+        },
+        "required": ["severity", "location", "claim", "evidence", "correction"]
+      }
+    },
+    "commands": {
+      "type": "array",
+      "maxItems": 20,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "command": {"type": "string", "minLength": 1},
+          "exit_code": {"type": "integer"},
+          "observation": {"type": "string", "minLength": 1}
+        },
+        "required": ["command", "exit_code", "observation"]
+      }
+    },
+    "blockers": {"type": "array", "items": {"type": "string"}, "maxItems": 20}
+  },
+  "required": ["verdict", "findings", "commands", "blockers"],
+  "allOf": [
+    {"if": {"properties": {"verdict": {"const": "pass"}}, "required": ["verdict"]}, "then": {"properties": {"findings": {"maxItems": 0}, "blockers": {"maxItems": 0}}}},
+    {"if": {"properties": {"verdict": {"const": "blocked"}}, "required": ["verdict"]}, "then": {"properties": {"blockers": {"minItems": 1}}}}
+  ]
+}

profiles/codex-harness-team/profile.toml

@@ -0,0 +1,105 @@
+schema_version = 8
+id = "codex-harness-team"
+version = "8.0.0"
+display_name = "Codex Harness Team"
+description = "A harness-first ChatGPT-only team combining fast native context isolation with a fresh mechanically supervised Sol critic."
+tags = ["engineering", "homogeneous-maker", "native", "hybrid", "featured"]
+maturity = "featured"
+root = "integrator"
+smoke = "smoke.toml"
+
+[agents.integrator]
+kind = "root"
+description = "Critical-path Sol engineer, sole writer, and final integrator"
+model = "codex_chatgpt_builtin__gpt_5_6_sol"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh", "max"]
+execution_mode = "goal"
+goal_token_budget = 600000
+max_goal_token_budget = 1000000
+stall_warning_seconds = 3600
+finalization_grace_seconds = 1800
+plan_reasoning = "high"
+permissions = "workspace-write"
+can_spawn = ["repo_scout", "invariant_designer", "fresh_critic"]
+max_active = 1
+trust = "high"
+verification = "risk_based"
+attachments_allowed = true
+requires_modalities = ["text", "image"]
+instructions = "agents/integrator.md"
+
+[agents.repo_scout]
+description = "Low-latency repository and test reconnaissance in a native context"
+model = "codex_chatgpt_builtin__gpt_5_6_luna"
+reasoning = "medium"
+allowed_reasoning_efforts = ["low", "medium", "high"]
+permissions = "read-only"
+backends = ["native"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["locate", "references", "test_analysis"]
+trust = "normal"
+verification = "material_changes"
+contract_enforcement = "warn"
+output_contract = "contracts/evidence.json"
+instructions = "agents/repo-scout.md"
+
+[agents.invariant_designer]
+description = "Independent native invariant and test designer"
+model = "codex_chatgpt_builtin__gpt_5_6_terra"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh", "max"]
+permissions = "read-only"
+backends = ["native"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["analysis", "test_design", "specification"]
+trust = "high"
+verification = "material_changes"
+contract_enforcement = "warn"
+output_contract = "contracts/evidence.json"
+instructions = "agents/invariant-designer.md"
+
+[agents.fresh_critic]
+description = "Fresh-context Sol critic after the root has an integrated diff"
+model = "codex_chatgpt_builtin__gpt_5_6_sol"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh", "max"]
+execution_mode = "turn"
+stall_warning_seconds = 1800
+finalization_grace_seconds = 900
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["review", "verification"]
+trust = "adversarial_reviewer"
+verification = "root_adjudication"
+contract_enforcement = "strict"
+output_contract = "contracts/review.json"
+instructions = "agents/fresh-critic.md"
+
+[agents.integrator.controls.repo_scout]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.integrator.controls.invariant_designer]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.integrator.controls.fresh_critic]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[coordination]
+mode = "harness_native_first"
+orchestration = "hybrid"
+max_active_agents = 4
+max_depth = 1
+max_children_per_agent = 3
+max_active_writers = 0
+native_max_concurrent_threads = 2
+native_nested_delegation = false
+reject_ancestor_role = true
+wait_policy = "dependency_only"
+write_conflict_policy = "reject"
+contradiction_policy = "primary_evidence"
+result_visibility = "ancestors"

profiles/codex-harness-team/smoke.toml

@@ -0,0 +1,32 @@
+schema_version = 8
+
+[[tasks]]
+agent = "integrator"
+task_kind = "analysis"
+task = "Inspect this profile read-only and explain what the native-first plus fresh-context critic topology demonstrates."
+mode = "read-only"
+wall_timeout_seconds = 600
+
+[[tasks]]
+agent = "repo_scout"
+task_kind = "locate"
+task = "Locate the native-agent snapshot generation code and return exact repository evidence in the required JSON shape."
+backend = "native"
+mode = "read-only"
+wall_timeout_seconds = 600
+
+[[tasks]]
+agent = "invariant_designer"
+task_kind = "test_design"
+task = "Derive two falsifying tests for the compiled profile concurrency limits and return the required evidence JSON."
+backend = "native"
+mode = "read-only"
+wall_timeout_seconds = 600
+
+[[tasks]]
+agent = "fresh_critic"
+task_kind = "review"
+task = "Review profiles/codex-harness-team/profile.toml for one concrete inconsistency. Return only the required review JSON with exact evidence."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900

profiles/competing-implementations-lab/README.md

@@ -0,0 +1,7 @@
+# Competing Implementations Lab
+
+This lab asks DeepSeek V4 Pro and Sonnet 5 to implement the same frozen contract independently and concurrently. Each worker writes only its isolated worktree; the Terra root keeps the canonical workspace unchanged until evidence selects one patch.
+
+Selection uses tests, benchmark results, maintainability evidence, patch review, and integration effort. Model identity and majority preference are inadmissible. The extra implementation is justified only when matched trials show enough quality gain to offset doubled calls and comparison work.
+
+Both candidates are independently hosted persistent Unix app-server workers with equal goal/turn, warning, and finalization policy. The judge can inspect, trace, steer, pause, continue, detach, finalize, fully stop, fork, or cancel either candidate without granting them control over one another. Client/host failure therefore cannot silently discard a candidate's history, evidence, or isolated patch.

profiles/competing-implementations-lab/agents/candidate.md

@@ -0,0 +1 @@
+Independently implement the exact supplied objective contract in your isolated worktree. Do not inspect or anticipate the competing candidate. Stay within scope, run the specified tests and benchmark, and return the contract hash, correlated commands, measurements, files, risks, and a minimal patch. The root alone selects and integrates.

profiles/competing-implementations-lab/agents/experiment-judge.md

@@ -0,0 +1 @@
+Define and hash one objective contract plus benchmark before launching both candidates concurrently. Give identical task context and disjoint isolated worktrees; keep the canonical workspace unchanged until selection. Read, explicitly accept or reject, and compare patches using tests, benchmark evidence, maintainability, scope, and integration effort. Do not vote, average outputs, prefer a maker, or merge both by default. Integrate only the selected patch after rechecking its base fingerprint.

profiles/competing-implementations-lab/contracts/candidate.json

@@ -0,0 +1,15 @@
+{
+  "title": "Matched isolated implementation candidate",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "contract_sha256": {"type": "string", "pattern": "^[0-9a-f]{64}$"},
+    "summary": {"type": "string", "minLength": 1},
+    "files": {"type": "array", "items": {"type": "string"}, "minItems": 1, "maxItems": 50, "uniqueItems": true},
+    "commands": {"type": "array", "minItems": 1, "maxItems": 30, "items": {"type": "object", "additionalProperties": false, "properties": {"command": {"type": "string", "minLength": 1}, "exit_code": {"type": "integer"}, "observation": {"type": "string", "minLength": 1}}, "required": ["command", "exit_code", "observation"]}},
+    "benchmark": {"type": "object", "additionalProperties": false, "properties": {"command": {"type": "string", "minLength": 1}, "exit_code": {"type": "integer"}, "measurement": {"type": "number", "minimum": 0}, "unit": {"type": "string", "minLength": 1}}, "required": ["command", "exit_code", "measurement", "unit"]},
+    "maintainability_evidence": {"type": "array", "items": {"type": "string"}, "maxItems": 30},
+    "risks": {"type": "array", "items": {"type": "string"}, "maxItems": 30}
+  },
+  "required": ["contract_sha256", "summary", "files", "commands", "benchmark", "maintainability_evidence", "risks"]
+}

profiles/competing-implementations-lab/profile.toml

@@ -0,0 +1,89 @@
+schema_version = 8
+id = "competing-implementations-lab"
+version = "8.0.0"
+display_name = "Competing Implementations Lab"
+description = "An experimental Terra root defines one objective contract while DeepSeek and Sonnet independently build competing isolated patches for evidence-based selection."
+tags = ["implementation", "competition", "isolated-worktrees", "lab"]
+maturity = "lab"
+root = "experiment_judge"
+smoke = "smoke.toml"
+
+[agents.experiment_judge]
+kind = "root"
+description = "Objective-contract author, benchmark owner, canonical writer, and selection authority"
+model = "codex_chatgpt_builtin__gpt_5_6_terra"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh", "max"]
+execution_mode = "goal"
+goal_token_budget = 400000
+max_goal_token_budget = 600000
+stall_warning_seconds = 3600
+finalization_grace_seconds = 1800
+plan_reasoning = "high"
+permissions = "workspace-write"
+can_spawn = ["deepseek_candidate", "sonnet_candidate"]
+max_active = 1
+trust = "high"
+verification = "always"
+instructions = "agents/experiment-judge.md"
+
+[agents.deepseek_candidate]
+description = "Independent DeepSeek implementation in an isolated worktree"
+model = "opencode_go_openai_chat__deepseek_v4_pro"
+reasoning = "high"
+allowed_reasoning_efforts = ["high", "max"]
+execution_mode = "goal"
+goal_token_budget = 300000
+max_goal_token_budget = 500000
+stall_warning_seconds = 1800
+finalization_grace_seconds = 1800
+permissions = "workspace-write"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["implement", "benchmark"]
+trust = "high"
+verification = "always"
+contract_enforcement = "strict"
+output_contract = "contracts/candidate.json"
+instructions = "agents/candidate.md"
+
+[agents.sonnet_candidate]
+description = "Independent Sonnet implementation in a separate isolated worktree"
+model = "opencode_zen_anthropic_messages__claude_sonnet_5"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh", "max"]
+execution_mode = "goal"
+goal_token_budget = 300000
+max_goal_token_budget = 500000
+stall_warning_seconds = 1800
+finalization_grace_seconds = 1800
+permissions = "workspace-write"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["implement", "benchmark"]
+trust = "high"
+verification = "always"
+contract_enforcement = "strict"
+output_contract = "contracts/candidate.json"
+instructions = "agents/candidate.md"
+
+[agents.experiment_judge.controls.deepseek_candidate]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.experiment_judge.controls.sonnet_candidate]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[coordination]
+mode = "matched_competing_implementations_experiment"
+orchestration = "mcp"
+max_active_agents = 3
+max_depth = 1
+max_children_per_agent = 2
+max_active_writers = 2
+reject_ancestor_role = true
+wait_policy = "dependency_only"
+write_conflict_policy = "reject"
+contradiction_policy = "primary_evidence"
+result_visibility = "session"

profiles/competing-implementations-lab/smoke.toml

@@ -0,0 +1,26 @@
+schema_version = 8
+
+[[tasks]]
+agent = "experiment_judge"
+task_kind = "analysis"
+task = "Inspect this lab read-only and explain how objective evidence selects one isolated candidate without voting."
+mode = "read-only"
+wall_timeout_seconds = 600
+
+[[tasks]]
+agent = "deepseek_candidate"
+task_kind = "implement"
+task = "Create only evals/competing-implementations/fixture/mmo-deepseek-smoke.txt containing the word candidate, verify it with a local command, and return candidate JSON using 64 zeroes as the supplied demonstration contract hash."
+backend = "mcp"
+mode = "workspace-write"
+write_scope = ["evals/competing-implementations/fixture/mmo-deepseek-smoke.txt"]
+wall_timeout_seconds = 900
+
+[[tasks]]
+agent = "sonnet_candidate"
+task_kind = "implement"
+task = "Create only evals/competing-implementations/fixture/mmo-sonnet-smoke.txt containing the word candidate, verify it with a local command, and return candidate JSON using 64 zeroes as the supplied demonstration contract hash."
+backend = "mcp"
+mode = "workspace-write"
+write_scope = ["evals/competing-implementations/fixture/mmo-sonnet-smoke.txt"]
+wall_timeout_seconds = 900

profiles/contract-first-refactoring/README.md

@@ -0,0 +1,7 @@
+# Contract-First Refactoring
+
+Use this featured profile for compatibility-sensitive migrations and refactors. Sonnet owns the immutable contract and production changes; MiniMax mines repository-wide constraints; Luna creates a test-only isolated patch; DeepSeek independently verifies the final result.
+
+The root freezes and hashes the normalized contract before downstream work. Test patches require explicit acceptance and integration, production source remains root-owned, and verification must cite the same contract hash. This differs from ordinary review by separating specification, tests, implementation, and compliance evidence.
+
+Every worker is a persistent supervised Unix app-server thread with strict retained results, goal/turn lifecycle, and warning-only stalls. The root controls all three; the compatibility verifier may steer or return the test author for correction but cannot accept or integrate that patch. Suspension preserves exact contract evidence, terminal history, trace, and the isolated test patch for deliberate continuation.

profiles/contract-first-refactoring/agents/compatibility-verifier.md

@@ -0,0 +1 @@
+Independently verify the final implementation against the exact frozen contract hash, existing tests, newly accepted tests, and backward-observable behavior. Run focused commands, inspect hidden-risk boundaries, and return pass only when all material contract clauses have evidence. Do not edit or silently revise the contract.

profiles/contract-first-refactoring/agents/contract-test-author.md

@@ -0,0 +1 @@
+Write tests only within the assigned test scope and only against the supplied frozen contract hash. Include tests that fail plausible incorrect implementations, boundary cases, and compatibility behavior; do not modify production source. Run focused commands and return correlated evidence plus a minimal isolated patch for explicit root acceptance and integration.

profiles/contract-first-refactoring/agents/invariant-miner.md

@@ -0,0 +1 @@
+Mine explicit and implicit compatibility constraints across the supplied repository corpus. Cite exact locations and distinguish guaranteed behavior, accidental behavior, and uncertainty. Do not edit or choose the implementation. Return a canonical contract payload and SHA-256 so downstream work can be tied to one immutable interpretation.

profiles/contract-first-refactoring/agents/refactor-lead.md

@@ -0,0 +1 @@
+Own the contract, production-source changes, and final integration. Launch invariant mining early while you inspect the code. Reconcile explicit specifications, tests, public surfaces, persisted data, and observable behavior into one normalized contract, then freeze and hash it before authorizing tests or implementation. Accept and explicitly integrate the test-only patch before changing source. Implement against the frozen contract, then request independent compatibility verification. If the contract changes, record a new hash and invalidate downstream evidence.

profiles/contract-first-refactoring/contracts/invariants.json

@@ -0,0 +1,27 @@
+{
+  "title": "Frozen compatibility contract",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "contract": {"type": "string", "minLength": 1},
+    "contract_sha256": {"type": "string", "pattern": "^[0-9a-f]{64}$"},
+    "clauses": {
+      "type": "array",
+      "minItems": 1,
+      "maxItems": 100,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "id": {"type": "string", "minLength": 1},
+          "invariant": {"type": "string", "minLength": 1},
+          "evidence_location": {"type": "string", "minLength": 1},
+          "classification": {"type": "string", "enum": ["guaranteed", "observed", "uncertain"]}
+        },
+        "required": ["id", "invariant", "evidence_location", "classification"]
+      }
+    },
+    "uncertainties": {"type": "array", "items": {"type": "string"}, "maxItems": 30}
+  },
+  "required": ["contract", "contract_sha256", "clauses", "uncertainties"]
+}

profiles/contract-first-refactoring/contracts/test-patch.json

@@ -0,0 +1,24 @@
+{
+  "title": "Contract-derived isolated test patch",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "contract_sha256": {"type": "string", "pattern": "^[0-9a-f]{64}$"},
+    "covered_clause_ids": {"type": "array", "items": {"type": "string"}, "minItems": 1, "maxItems": 100, "uniqueItems": true},
+    "test_files": {"type": "array", "items": {"type": "string"}, "minItems": 1, "maxItems": 30, "uniqueItems": true},
+    "plausible_failures_caught": {"type": "array", "items": {"type": "string"}, "minItems": 1, "maxItems": 30},
+    "commands": {
+      "type": "array",
+      "minItems": 1,
+      "maxItems": 30,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {"command": {"type": "string", "minLength": 1}, "exit_code": {"type": "integer"}, "observation": {"type": "string", "minLength": 1}},
+        "required": ["command", "exit_code", "observation"]
+      }
+    },
+    "limitations": {"type": "array", "items": {"type": "string"}, "maxItems": 20}
+  },
+  "required": ["contract_sha256", "covered_clause_ids", "test_files", "plausible_failures_caught", "commands", "limitations"]
+}

profiles/contract-first-refactoring/contracts/verification.json

@@ -0,0 +1,28 @@
+{
+  "title": "Frozen-contract compatibility verification",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "verdict": {"type": "string", "enum": ["pass", "fail", "blocked"]},
+    "contract_sha256": {"type": "string", "pattern": "^[0-9a-f]{64}$"},
+    "verified_clause_ids": {"type": "array", "items": {"type": "string"}, "maxItems": 100, "uniqueItems": true},
+    "commands": {
+      "type": "array",
+      "maxItems": 30,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {"command": {"type": "string", "minLength": 1}, "exit_code": {"type": "integer"}, "observation": {"type": "string", "minLength": 1}},
+        "required": ["command", "exit_code", "observation"]
+      }
+    },
+    "findings": {"type": "array", "items": {"type": "string"}, "maxItems": 30},
+    "blockers": {"type": "array", "items": {"type": "string"}, "maxItems": 20}
+  },
+  "required": ["verdict", "contract_sha256", "verified_clause_ids", "commands", "findings", "blockers"],
+  "allOf": [
+    {"if": {"properties": {"verdict": {"const": "pass"}}, "required": ["verdict"]}, "then": {"properties": {"verified_clause_ids": {"minItems": 1}, "commands": {"minItems": 1}, "findings": {"maxItems": 0}, "blockers": {"maxItems": 0}}}},
+    {"if": {"properties": {"verdict": {"const": "fail"}}, "required": ["verdict"]}, "then": {"properties": {"findings": {"minItems": 1}}}},
+    {"if": {"properties": {"verdict": {"const": "blocked"}}, "required": ["verdict"]}, "then": {"properties": {"blockers": {"minItems": 1}}}}
+  ]
+}

profiles/contract-first-refactoring/profile.toml

@@ -0,0 +1,114 @@
+schema_version = 8
+id = "contract-first-refactoring"
+version = "8.0.0"
+display_name = "Contract-First Refactoring"
+description = "A Sonnet root freezes compatibility invariants before implementation, accepts an independent test patch, and closes with economical regression verification."
+tags = ["refactoring", "compatibility", "test-first", "featured"]
+maturity = "featured"
+root = "refactor_lead"
+smoke = "smoke.toml"
+
+[agents.refactor_lead]
+kind = "root"
+description = "Contract owner, sole source writer, and final refactor integrator"
+model = "opencode_zen_anthropic_messages__claude_sonnet_5"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh", "max"]
+execution_mode = "goal"
+goal_token_budget = 600000
+max_goal_token_budget = 1000000
+stall_warning_seconds = 3600
+finalization_grace_seconds = 1800
+plan_reasoning = "high"
+permissions = "workspace-write"
+can_spawn = ["invariant_miner", "contract_test_author", "compatibility_verifier"]
+max_active = 1
+trust = "high"
+verification = "always"
+instructions = "agents/refactor-lead.md"
+
+[agents.invariant_miner]
+description = "Long-context read-only miner of explicit and implicit compatibility constraints"
+model = "opencode_go_anthropic_messages__minimax_m3"
+reasoning = "none"
+allowed_reasoning_efforts = ["none"]
+execution_mode = "goal"
+goal_token_budget = 250000
+max_goal_token_budget = 400000
+stall_warning_seconds = 1800
+finalization_grace_seconds = 900
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["invariant_analysis", "compatibility_analysis"]
+trust = "high"
+verification = "root_adjudication"
+contract_enforcement = "strict"
+output_contract = "contracts/invariants.json"
+instructions = "agents/invariant-miner.md"
+
+[agents.contract_test_author]
+description = "Independent Luna author of contract-derived tests in a test-only worktree"
+model = "codex_chatgpt_builtin__gpt_5_6_luna"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh"]
+execution_mode = "goal"
+goal_token_budget = 250000
+max_goal_token_budget = 400000
+stall_warning_seconds = 1800
+finalization_grace_seconds = 1200
+permissions = "workspace-write"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["test"]
+trust = "high"
+verification = "always"
+contract_enforcement = "strict"
+output_contract = "contracts/test-patch.json"
+instructions = "agents/contract-test-author.md"
+
+[agents.compatibility_verifier]
+description = "Independent economical compatibility and regression verifier"
+model = "opencode_go_openai_chat__deepseek_v4_pro"
+reasoning = "high"
+allowed_reasoning_efforts = ["high", "max"]
+execution_mode = "turn"
+stall_warning_seconds = 1800
+finalization_grace_seconds = 900
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["verification", "test_analysis"]
+trust = "adversarial_reviewer"
+verification = "root_adjudication"
+contract_enforcement = "strict"
+output_contract = "contracts/verification.json"
+instructions = "agents/compatibility-verifier.md"
+
+[agents.refactor_lead.controls.invariant_miner]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.refactor_lead.controls.contract_test_author]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.refactor_lead.controls.compatibility_verifier]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.compatibility_verifier.controls.contract_test_author]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[coordination]
+mode = "contract_test_refactor_verify"
+orchestration = "mcp"
+max_active_agents = 3
+max_depth = 1
+max_children_per_agent = 3
+max_active_writers = 1
+reject_ancestor_role = true
+wait_policy = "dependency_only"
+write_conflict_policy = "reject"
+contradiction_policy = "primary_evidence"
+result_visibility = "ancestors"

profiles/contract-first-refactoring/smoke.toml

@@ -0,0 +1,32 @@
+schema_version = 8
+
+[[tasks]]
+agent = "refactor_lead"
+task_kind = "analysis"
+task = "Inspect the profile compiler read-only and explain why the contract hash must precede test and implementation work."
+mode = "read-only"
+wall_timeout_seconds = 600
+
+[[tasks]]
+agent = "invariant_miner"
+task_kind = "invariant_analysis"
+task = "Mine two compatibility invariants from libexec/mmo_profiles.py. Return only the frozen-contract JSON with a SHA-256 matching the exact contract string."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900
+
+[[tasks]]
+agent = "contract_test_author"
+task_kind = "test"
+task = "Do not edit files. Describe one contract-derived test for rejecting schema v1 and return the test-patch JSON using 64 zeroes as the supplied demonstration contract hash and correlated command evidence."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900
+
+[[tasks]]
+agent = "compatibility_verifier"
+task_kind = "verification"
+task = "Verify read-only that the profile compiler rejects schema v1. Return only the verification JSON using 64 zeroes as the supplied demonstration contract hash and captured command evidence."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900

profiles/high-confidence-debugging/README.md

@@ -0,0 +1,7 @@
+# High-Confidence Debugging
+
+Use this featured profile for ambiguous or high-impact failures where changing code before independent reproduction would be risky. The DeepSeek V4 Pro root investigates and repairs; Sonnet independently reproduces and may contribute only a test-scope patch; fresh Sol verification attacks the fix, with one correction/reverification cycle.
+
+Only one worker runs alongside the root at a time. The sequence deliberately overlaps root diagnosis with reproduction, then enforces reproduction, repair, and verification barriers. Missing optional routes reduce confidence but do not prevent the root from reporting a blocked or evidence-limited diagnosis.
+
+Reproducer and verifier are persistent supervised Unix app-server workers. The root controls both, and each can steer or return the other to the precise missing proof through reciprocal action grants without gaining patch-integration authority. Long inference uses generous goal tokens or a clock-free turn; silence only warns, and suspension retains terminal history, reproduction artifacts, traces, and partial conclusions.

profiles/high-confidence-debugging/agents/debug-lead.md

@@ -0,0 +1 @@
+Own diagnosis, canonical repair, and final correctness. Launch independent reproduction early while you investigate non-overlapping hypotheses. Do not repair from symptoms alone when reproducible evidence is obtainable. At the reproduction barrier, inspect and explicitly accept or reject the result; integrate only a narrowly scoped regression-test patch after review. Implement the repair yourself, then request fresh verification. One correction and reverification cycle is available. Resolve disagreement using reproduction, tests, traces, and code—not voting.

profiles/high-confidence-debugging/agents/fix-verifier.md

@@ -0,0 +1 @@
+Adversarially verify the proposed fix against the original reproduction, hidden adjacent cases, and plausible regressions. Run focused commands and correlate their exit codes. Do not edit. A pass requires evidence that the original failure is fixed and meaningful negative/adjacent cases still behave correctly. Identify a precise correction target for every failure.

profiles/high-confidence-debugging/agents/independent-reproducer.md

@@ -0,0 +1 @@
+Independently reproduce the reported failure before seeing a proposed repair. Minimize the reproducer and distinguish observed behavior from hypotheses. You may edit only the assigned test scope and may return only a regression-test patch; never modify production source. Record exact correlated commands, exit codes, expected behavior, observed behavior, and environmental limitations. The root alone accepts and integrates your patch.

profiles/high-confidence-debugging/contracts/reproduction.json

@@ -0,0 +1,33 @@
+{
+  "title": "Independent failure reproduction",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "status": {"type": "string", "enum": ["reproduced", "not_reproduced", "blocked"]},
+    "minimal_reproduction": {"type": "string"},
+    "expected": {"type": "string", "minLength": 1},
+    "observed": {"type": "string", "minLength": 1},
+    "commands": {
+      "type": "array",
+      "maxItems": 30,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "command": {"type": "string", "minLength": 1},
+          "exit_code": {"type": "integer"},
+          "observation": {"type": "string", "minLength": 1}
+        },
+        "required": ["command", "exit_code", "observation"]
+      }
+    },
+    "test_files": {"type": "array", "items": {"type": "string"}, "maxItems": 20, "uniqueItems": true},
+    "hypotheses": {"type": "array", "items": {"type": "string"}, "maxItems": 20},
+    "limitations": {"type": "array", "items": {"type": "string"}, "maxItems": 20}
+  },
+  "required": ["status", "minimal_reproduction", "expected", "observed", "commands", "test_files", "hypotheses", "limitations"],
+  "allOf": [
+    {"if": {"properties": {"status": {"const": "reproduced"}}, "required": ["status"]}, "then": {"properties": {"minimal_reproduction": {"minLength": 1}, "commands": {"minItems": 1}, "test_files": {"minItems": 1}}}},
+    {"if": {"properties": {"status": {"const": "blocked"}}, "required": ["status"]}, "then": {"properties": {"limitations": {"minItems": 1}}}}
+  ]
+}

profiles/high-confidence-debugging/contracts/verification.json

@@ -0,0 +1,33 @@
+{
+  "title": "Adversarial fix verification",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "verdict": {"type": "string", "enum": ["pass", "fail", "blocked"]},
+    "original_failure_fixed": {"type": "boolean"},
+    "adjacent_cases_preserved": {"type": "boolean"},
+    "commands": {
+      "type": "array",
+      "maxItems": 30,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "command": {"type": "string", "minLength": 1},
+          "exit_code": {"type": "integer"},
+          "observation": {"type": "string", "minLength": 1}
+        },
+        "required": ["command", "exit_code", "observation"]
+      }
+    },
+    "findings": {"type": "array", "items": {"type": "string"}, "maxItems": 30},
+    "correction_targets": {"type": "array", "items": {"type": "string"}, "maxItems": 30},
+    "blockers": {"type": "array", "items": {"type": "string"}, "maxItems": 20}
+  },
+  "required": ["verdict", "original_failure_fixed", "adjacent_cases_preserved", "commands", "findings", "correction_targets", "blockers"],
+  "allOf": [
+    {"if": {"properties": {"verdict": {"const": "pass"}}, "required": ["verdict"]}, "then": {"properties": {"original_failure_fixed": {"const": true}, "adjacent_cases_preserved": {"const": true}, "commands": {"minItems": 1}, "findings": {"maxItems": 0}, "blockers": {"maxItems": 0}}}},
+    {"if": {"properties": {"verdict": {"const": "fail"}}, "required": ["verdict"]}, "then": {"properties": {"findings": {"minItems": 1}, "correction_targets": {"minItems": 1}}}},
+    {"if": {"properties": {"verdict": {"const": "blocked"}}, "required": ["verdict"]}, "then": {"properties": {"blockers": {"minItems": 1}}}}
+  ]
+}

profiles/high-confidence-debugging/profile.toml

@@ -0,0 +1,93 @@
+schema_version = 8
+id = "high-confidence-debugging"
+version = "8.0.0"
+display_name = "High-Confidence Debugging"
+description = "An economical DeepSeek diagnostician overlaps an independent Sonnet reproduction, repairs only after reproduction evidence, and submits the fix to fresh Sol verification."
+tags = ["debugging", "reproduction", "actor-critic", "featured"]
+maturity = "featured"
+root = "debug_lead"
+smoke = "smoke.toml"
+
+[agents.debug_lead]
+kind = "root"
+description = "Diagnostician, sole repair writer, correction owner, and final debugger"
+model = "opencode_go_openai_chat__deepseek_v4_pro"
+reasoning = "high"
+allowed_reasoning_efforts = ["high", "max"]
+execution_mode = "goal"
+goal_token_budget = 600000
+max_goal_token_budget = 1000000
+stall_warning_seconds = 3600
+finalization_grace_seconds = 1800
+plan_reasoning = "high"
+permissions = "workspace-write"
+can_spawn = ["independent_reproducer", "fix_verifier"]
+max_active = 1
+trust = "high"
+verification = "always"
+instructions = "agents/debug-lead.md"
+
+[agents.independent_reproducer]
+description = "Independent reproducer allowed to return only a regression-test patch"
+model = "opencode_zen_anthropic_messages__claude_sonnet_5"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh", "max"]
+execution_mode = "goal"
+goal_token_budget = 200000
+max_goal_token_budget = 300000
+stall_warning_seconds = 1800
+finalization_grace_seconds = 1200
+permissions = "workspace-write"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["reproduction", "test"]
+trust = "high"
+verification = "always"
+contract_enforcement = "strict"
+output_contract = "contracts/reproduction.json"
+instructions = "agents/independent-reproducer.md"
+
+[agents.fix_verifier]
+description = "Fresh Sol adversarial verifier, reusable once for a correction/reverification cycle"
+model = "codex_chatgpt_builtin__gpt_5_6_sol"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh", "max"]
+execution_mode = "turn"
+stall_warning_seconds = 1800
+finalization_grace_seconds = 900
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["verification", "review", "test_analysis"]
+trust = "adversarial_reviewer"
+verification = "root_adjudication"
+contract_enforcement = "strict"
+output_contract = "contracts/verification.json"
+instructions = "agents/fix-verifier.md"
+
+[agents.debug_lead.controls.independent_reproducer]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.debug_lead.controls.fix_verifier]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.independent_reproducer.controls.fix_verifier]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.fix_verifier.controls.independent_reproducer]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[coordination]
+mode = "reproduce_repair_verify"
+orchestration = "mcp"
+max_active_agents = 2
+max_depth = 1
+max_children_per_agent = 2
+max_active_writers = 1
+reject_ancestor_role = true
+wait_policy = "dependency_only"
+write_conflict_policy = "reject"
+contradiction_policy = "primary_evidence"
+result_visibility = "ancestors"

profiles/high-confidence-debugging/smoke.toml

@@ -0,0 +1,24 @@
+schema_version = 8
+
+[[tasks]]
+agent = "debug_lead"
+task_kind = "analysis"
+task = "Inspect the debugging evaluation fixture read-only and describe the reproduction, repair, and verification barriers."
+mode = "read-only"
+wall_timeout_seconds = 600
+
+[[tasks]]
+agent = "independent_reproducer"
+task_kind = "reproduction"
+task = "Reproduce the currently passing debugging fixture without editing production source. If no failure exists, return only valid reproduction JSON with status not_reproduced and correlated command evidence; make no changes."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900
+
+[[tasks]]
+agent = "fix_verifier"
+task_kind = "verification"
+task = "Verify the current debugging fixture tests read-only. Return only valid verification JSON with captured command evidence and no unsupported pass claim."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900

profiles/incident-hypothesis-triage/README.md

@@ -0,0 +1,7 @@
+# Incident Hypothesis Triage
+
+Use this featured profile for read-only production incident triage or difficult causal investigations. GLM-5.3 on the Z.AI Coding Plan synthesizes evidence; fast DeepSeek Flash performs bounded local collection; Nemotron through a pinned OpenRouter policy develops and attacks alternatives.
+
+The topology is intentionally heterogeneous and read-only. It may assert root cause only from a reproducible evidence chain; otherwise it reports ranked unresolved hypotheses and safe next probes. OpenRouter policy and serving telemetry make route claims auditable, but do not prove semantic independence or correctness.
+
+Both workers are persistent supervised Unix app-server threads. The root controls each, and evidence runner and causal challenger can steer, interrupt, pause, continue, detach, finalize, fully stop, or cancel one another through reciprocal action grants when a claim needs a sharper test. Goal tokens or clock-free turns accommodate slow providers while retained terminal history, traces, and partial evidence prevent transport loss from deleting the investigation.

profiles/incident-hypothesis-triage/agents/causal-challenger.md

@@ -0,0 +1 @@
+Independently develop alternative causal hypotheses and try to falsify each using supplied evidence. Separate observation, inference, and speculation. Give predictions that would distinguish hypotheses and evidence still needed. Never assert a root cause by confidence, consensus, or model reputation; an OpenRouter route policy reduces routing ambiguity but does not itself prove semantic independence.

profiles/incident-hypothesis-triage/agents/evidence-runner.md

@@ -0,0 +1 @@
+Collect or correlate only the bounded local evidence requested. Record exact commands, exit codes, timestamps, paths, hashes, and observations; do not change files, access the network, or decide root cause. Identify gaps and contradictions. A command claim must be present in captured tool events.

profiles/incident-hypothesis-triage/agents/incident-lead.md

@@ -0,0 +1 @@
+Remain read-only. Own the incident timeline, evidence graph, and final adjudication. Launch bounded evidence collection and the independent causal challenger early, then continue correlating supplied evidence. Read and explicitly accept or reject worker results. A root cause may be asserted only when a reproducible evidence chain distinguishes it from alternatives; otherwise publish ranked unresolved hypotheses, missing observations, and safe next probes. Treat provider failure or an unavailable challenger as visible uncertainty, not permission to invent certainty.

profiles/incident-hypothesis-triage/contracts/evidence.json

@@ -0,0 +1,37 @@
+{
+  "title": "Bounded incident evidence",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "question": {"type": "string", "minLength": 1},
+    "observations": {
+      "type": "array",
+      "maxItems": 100,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "id": {"type": "string", "minLength": 1},
+          "timestamp": {"type": ["string", "null"]},
+          "source": {"type": "string", "minLength": 1},
+          "sha256": {"type": ["string", "null"], "pattern": "^[0-9a-f]{64}$"},
+          "observation": {"type": "string", "minLength": 1}
+        },
+        "required": ["id", "timestamp", "source", "sha256", "observation"]
+      }
+    },
+    "commands": {
+      "type": "array",
+      "maxItems": 30,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {"command": {"type": "string", "minLength": 1}, "exit_code": {"type": "integer"}, "observation": {"type": "string", "minLength": 1}},
+        "required": ["command", "exit_code", "observation"]
+      }
+    },
+    "contradictions": {"type": "array", "items": {"type": "string"}, "maxItems": 30},
+    "gaps": {"type": "array", "items": {"type": "string"}, "maxItems": 30}
+  },
+  "required": ["question", "observations", "commands", "contradictions", "gaps"]
+}

profiles/incident-hypothesis-triage/contracts/hypotheses.json

@@ -0,0 +1,28 @@
+{
+  "title": "Independent causal hypotheses",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "hypotheses": {
+      "type": "array",
+      "minItems": 2,
+      "maxItems": 20,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "id": {"type": "string", "minLength": 1},
+          "claim": {"type": "string", "minLength": 1},
+          "supporting_evidence_ids": {"type": "array", "items": {"type": "string"}, "maxItems": 30, "uniqueItems": true},
+          "contradicting_evidence_ids": {"type": "array", "items": {"type": "string"}, "maxItems": 30, "uniqueItems": true},
+          "distinguishing_prediction": {"type": "string", "minLength": 1},
+          "status": {"type": "string", "enum": ["supported", "weakened", "unresolved"]}
+        },
+        "required": ["id", "claim", "supporting_evidence_ids", "contradicting_evidence_ids", "distinguishing_prediction", "status"]
+      }
+    },
+    "missing_evidence": {"type": "array", "items": {"type": "string"}, "maxItems": 30},
+    "root_cause_asserted": {"const": false}
+  },
+  "required": ["hypotheses", "missing_evidence", "root_cause_asserted"]
+}

profiles/incident-hypothesis-triage/profile.toml

@@ -0,0 +1,99 @@
+schema_version = 8
+id = "incident-hypothesis-triage"
+version = "8.0.0"
+display_name = "Incident Hypothesis Triage"
+description = "A read-only GLM incident lead combines bounded DeepSeek evidence collection with a separately routed Nemotron causal challenger."
+tags = ["incident", "causal-analysis", "provider-diversity", "read-only", "featured"]
+maturity = "featured"
+root = "incident_lead"
+smoke = "smoke.toml"
+
+[agents.incident_lead]
+kind = "root"
+description = "Read-only evidence synthesizer and final hypothesis adjudicator"
+model = "zai_coding_responses__glm_5_3"
+reasoning = "max"
+allowed_reasoning_efforts = ["high", "max"]
+execution_mode = "goal"
+goal_token_budget = 300000
+max_goal_token_budget = 500000
+stall_warning_seconds = 3600
+finalization_grace_seconds = 1200
+plan_reasoning = "max"
+permissions = "read-only"
+can_spawn = ["evidence_runner", "causal_challenger"]
+max_active = 1
+trust = "high"
+verification = "always"
+network_access = false
+web_search = "disabled"
+instructions = "agents/incident-lead.md"
+
+[agents.evidence_runner]
+description = "Fast bounded evidence collector and correlator, reusable for one follow-up query"
+model = "opencode_go_openai_chat__deepseek_v4_flash"
+reasoning = "high"
+allowed_reasoning_efforts = ["low", "high", "max"]
+execution_mode = "turn"
+stall_warning_seconds = 1800
+finalization_grace_seconds = 1200
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = []
+max_active = 2
+allowed_task_kinds = ["evidence_collection", "correlation"]
+trust = "high"
+verification = "always"
+network_access = false
+web_search = "disabled"
+contract_enforcement = "strict"
+output_contract = "contracts/evidence.json"
+instructions = "agents/evidence-runner.md"
+
+[agents.causal_challenger]
+description = "Independent causal analyst that develops and attacks alternative hypotheses"
+model = "openrouter_openai_chat__nvidia_nemotron_3_ultra_550b_a55b"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high"]
+execution_mode = "goal"
+goal_token_budget = 150000
+max_goal_token_budget = 250000
+stall_warning_seconds = 1800
+finalization_grace_seconds = 900
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["causal_analysis", "hypothesis_review"]
+trust = "adversarial_reviewer"
+verification = "root_adjudication"
+network_access = false
+web_search = "disabled"
+contract_enforcement = "strict"
+output_contract = "contracts/hypotheses.json"
+instructions = "agents/causal-challenger.md"
+
+[agents.incident_lead.controls.evidence_runner]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.incident_lead.controls.causal_challenger]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.evidence_runner.controls.causal_challenger]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.causal_challenger.controls.evidence_runner]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[coordination]
+mode = "evidence_gated_incident_triage"
+orchestration = "mcp"
+max_active_agents = 3
+max_depth = 1
+max_children_per_agent = 2
+max_active_writers = 0
+reject_ancestor_role = true
+wait_policy = "dependency_only"
+write_conflict_policy = "reject"
+contradiction_policy = "primary_evidence"
+result_visibility = "ancestors"

profiles/incident-hypothesis-triage/smoke.toml

@@ -0,0 +1,24 @@
+schema_version = 8
+
+[[tasks]]
+agent = "incident_lead"
+task_kind = "analysis"
+task = "Inspect the incident profile read-only and explain the evidence threshold for asserting root cause."
+mode = "read-only"
+wall_timeout_seconds = 600
+
+[[tasks]]
+agent = "evidence_runner"
+task_kind = "evidence_collection"
+task = "Collect bounded evidence about the current catalog schema version using local commands only. Return only the required incident-evidence JSON with correlated commands."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900
+
+[[tasks]]
+agent = "causal_challenger"
+task_kind = "causal_analysis"
+task = "Given only the fact that a profile compile failed, produce at least two unresolved alternative hypotheses and distinguishing predictions. Return only the required hypotheses JSON and do not assert a root cause."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900

profiles/research-backed-engineering/README.md

@@ -0,0 +1,15 @@
+# Research-Backed Engineering
+
+Use this profile when implementation correctness depends on current external APIs, standards, advisories, provider behavior, or versioned specifications.
+
+```text
+MiniMax M3 research lead (active root, sole writer, built-in live web search)
+├── Luna source researcher (MCP, read-only, built-in live web search)
+└── Claude Sonnet 5 source auditor (MCP, read-only, built-in live web search)
+```
+
+Independent research can run in parallel while the root maps the repository, but an authoritative dated dossier is a real dependency before root-owned implementation. The auditor independently re-fetches pivotal sources and checks both the dossier and change. Exact models, live-search grants, contracts, goal/turn lifecycle, and sole-writer boundary are mechanical. Source authority and implementation ordering are instruction-driven and evaluated.
+
+Cost and latency are high. Conflicting or unavailable primary sources produce a blocked or reduced-confidence result. Operators may clone the profile and grant a registered Firecrawl-like Tool MCP server only to the researcher; no deployment-specific server is bundled.
+
+Researcher and auditor are persistent supervised Unix app-server workers. The root controls both; the auditor may steer, interrupt, pause, continue, detach, or return the researcher for a source correction while acceptance remains with normal lineage authority. Pending input remains durable, and transport suspension retains terminal history, the dated dossier, trace, and partial source evidence.

profiles/research-backed-engineering/agents/research-lead.md

@@ -0,0 +1,3 @@
+Own the repository work, canonical writes, and final source adjudication. Split independent current-source questions early and continue inspecting the repository while researchers browse. Require every consequential claim to carry a URL, publication or update date, retrieval timestamp, authority classification, and concise evidence. Re-fetch pivotal primary sources yourself before allowing them to govern implementation. Use the source auditor when sources conflict, the change is temporally sensitive, or source-to-code compliance is consequential.
+
+Contracts validate shape and captured-event correlation, not truth. Resolve conflicts by authority, date, direct applicability, and reproducible behavior. Explicitly accept or reject dossiers; record unavailable routes and unresolved uncertainty. Never treat a search snippet or worker consensus as evidence.

profiles/research-backed-engineering/agents/source-auditor.md

@@ -0,0 +1 @@
+Independently audit the source dossier and resulting code. Revisit primary sources when permitted, check dates and applicability, identify contradictions or unsupported leaps, and map material requirements to exact implementation evidence. A valid JSON result is not proof of truth. Return a pass only when pivotal claims and code are independently supported; otherwise identify the precise re-fetch or empirical test needed.

profiles/research-backed-engineering/agents/source-researcher.md

@@ -0,0 +1 @@
+Research one bounded question using current primary sources first. Return URLs, source dates, retrieval timestamps, authority, directly supported claims, conflicts, and limitations. Separate source statements from your inference and keep excerpts short. Do not edit the repository or expand into the other researcher's question.

profiles/research-backed-engineering/contracts/research.json

@@ -0,0 +1,44 @@
+{
+  "title": "Dated current-source dossier",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "status": {"type": "string", "enum": ["supported", "conflict", "blocked"]},
+    "question": {"type": "string", "minLength": 1},
+    "claims": {
+      "type": "array",
+      "maxItems": 50,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "claim": {"type": "string", "minLength": 1},
+          "source_url": {"type": "string", "format": "uri"},
+          "source_date": {"type": "string", "format": "date"},
+          "retrieved_at": {"type": "string", "format": "date-time"},
+          "authority": {"type": "string", "enum": ["primary", "supporting"]},
+          "evidence": {"type": "string", "minLength": 1},
+          "inference": {"type": "boolean"}
+        },
+        "required": ["claim", "source_url", "source_date", "retrieved_at", "authority", "evidence", "inference"]
+      }
+    },
+    "conflicts": {"type": "array", "items": {"type": "string"}, "maxItems": 30},
+    "limitations": {"type": "array", "items": {"type": "string"}, "maxItems": 30}
+  },
+  "required": ["status", "question", "claims", "conflicts", "limitations"],
+  "allOf": [
+    {
+      "if": {"properties": {"status": {"const": "supported"}}, "required": ["status"]},
+      "then": {"properties": {"claims": {"minItems": 1}, "conflicts": {"maxItems": 0}}}
+    },
+    {
+      "if": {"properties": {"status": {"const": "conflict"}}, "required": ["status"]},
+      "then": {"properties": {"conflicts": {"minItems": 1}}}
+    },
+    {
+      "if": {"properties": {"status": {"const": "blocked"}}, "required": ["status"]},
+      "then": {"properties": {"limitations": {"minItems": 1}}}
+    }
+  ]
+}

profiles/research-backed-engineering/contracts/review.json

@@ -0,0 +1,37 @@
+{
+  "title": "Independent source-to-code audit",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "verdict": {"type": "string", "enum": ["pass", "concerns", "fail", "blocked"]},
+    "checked_sources": {"type": "array", "items": {"type": "string", "format": "uri"}, "maxItems": 50, "uniqueItems": true},
+    "findings": {
+      "type": "array",
+      "maxItems": 50,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "claim": {"type": "string", "minLength": 1},
+          "source_url": {"type": "string", "format": "uri"},
+          "code_location": {"type": "string", "minLength": 1},
+          "evidence": {"type": "string", "minLength": 1}
+        },
+        "required": ["claim", "source_url", "code_location", "evidence"]
+      }
+    },
+    "conflicts": {"type": "array", "items": {"type": "string"}, "maxItems": 30},
+    "blockers": {"type": "array", "items": {"type": "string"}, "maxItems": 30}
+  },
+  "required": ["verdict", "checked_sources", "findings", "conflicts", "blockers"],
+  "allOf": [
+    {
+      "if": {"properties": {"verdict": {"const": "pass"}}, "required": ["verdict"]},
+      "then": {"properties": {"checked_sources": {"minItems": 1}, "findings": {"maxItems": 0}, "blockers": {"maxItems": 0}}}
+    },
+    {
+      "if": {"properties": {"verdict": {"const": "blocked"}}, "required": ["verdict"]},
+      "then": {"properties": {"blockers": {"minItems": 1}}}
+    }
+  ]
+}

profiles/research-backed-engineering/profile.toml

@@ -0,0 +1,95 @@
+schema_version = 8
+id = "research-backed-engineering"
+version = "8.0.0"
+display_name = "Research-Backed Engineering"
+description = "A long-context MiniMax root integrates current primary-source research, re-fetches pivotal claims, and invites an independent source-to-code audit."
+tags = ["research", "engineering", "current-sources", "featured"]
+maturity = "featured"
+root = "research_lead"
+smoke = "smoke.toml"
+
+[agents.research_lead]
+kind = "root"
+description = "Long-context source adjudicator, repository engineer, sole writer, and final authority"
+model = "opencode_go_anthropic_messages__minimax_m3"
+reasoning = "none"
+allowed_reasoning_efforts = ["none"]
+execution_mode = "goal"
+goal_token_budget = 600000
+max_goal_token_budget = 1000000
+stall_warning_seconds = 3600
+finalization_grace_seconds = 1800
+permissions = "workspace-write"
+can_spawn = ["source_researcher", "source_auditor"]
+max_active = 1
+trust = "high"
+verification = "always"
+network_access = false
+web_search = "live"
+instructions = "agents/research-lead.md"
+
+[agents.source_researcher]
+description = "Fast independent live-web researcher for one bounded current-source question"
+model = "codex_chatgpt_builtin__gpt_5_6_luna"
+reasoning = "medium"
+allowed_reasoning_efforts = ["low", "medium", "high"]
+execution_mode = "goal"
+goal_token_budget = 200000
+max_goal_token_budget = 300000
+stall_warning_seconds = 1800
+finalization_grace_seconds = 900
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = []
+max_active = 2
+allowed_task_kinds = ["research", "source_verification"]
+trust = "high"
+verification = "always"
+network_access = false
+web_search = "live"
+contract_enforcement = "strict"
+output_contract = "contracts/research.json"
+instructions = "agents/source-researcher.md"
+
+[agents.source_auditor]
+description = "Conditional independent audit of authority, currency, conflicts, and source-to-code compliance"
+model = "opencode_zen_anthropic_messages__claude_sonnet_5"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh", "max"]
+execution_mode = "turn"
+stall_warning_seconds = 1800
+finalization_grace_seconds = 900
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["source_audit", "review", "verification"]
+trust = "adversarial_reviewer"
+verification = "root_adjudication"
+network_access = false
+web_search = "live"
+contract_enforcement = "strict"
+output_contract = "contracts/review.json"
+instructions = "agents/source-auditor.md"
+
+[agents.research_lead.controls.source_researcher]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.research_lead.controls.source_auditor]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.source_auditor.controls.source_researcher]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[coordination]
+mode = "research_implement_verify"
+orchestration = "mcp"
+max_active_agents = 3
+max_depth = 1
+max_children_per_agent = 2
+max_active_writers = 0
+reject_ancestor_role = true
+wait_policy = "dependency_only"
+write_conflict_policy = "reject"
+contradiction_policy = "primary_evidence"
+result_visibility = "ancestors"

profiles/research-backed-engineering/smoke.toml

@@ -0,0 +1,24 @@
+schema_version = 8
+
+[[tasks]]
+agent = "research_lead"
+task_kind = "analysis"
+task = "Inspect the repository read-only and identify the mechanical gates that distinguish current-source evidence from unsupported claims."
+mode = "read-only"
+wall_timeout_seconds = 600
+
+[[tasks]]
+agent = "source_researcher"
+task_kind = "research"
+task = "Using a current primary source, verify one documented Python language behavior relevant to this repository. Return only the dated dossier JSON and separate inference from source evidence."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900
+
+[[tasks]]
+agent = "source_auditor"
+task_kind = "source_audit"
+task = "Audit docs/EXTERNAL-VERACITY.md against one directly cited primary source. Return only the source-to-code audit JSON, or blocked with a precise reason."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900

profiles/route-resilience-lab/README.md

@@ -0,0 +1,7 @@
+# Route Resilience Lab
+
+This lab holds the DeepSeek V4 Pro model family and prompt constant while varying access infrastructure: OpenCode Go versus OpenRouter pinned to DeepSeek, with fallback disabled. Run the arms as a deliberate shadow pair or select one after a typed credential-loss, rate-limit, or timeout fault.
+
+It does not implement transparent automatic failover and does not infer the serving provider from model output. Session availability, requested policy, gateway retries, fallback index, actual serving slug, latency, and build fingerprint are the evidence. Duplicate calls are intentional only in matched drift trials.
+
+Each route arm is a separate persistent supervised Unix app-server worker with the same goal/turn and warning policy. The observer can inspect, trace, steer, interrupt, pause, continue, detach, finalize, fully stop, fork, or cancel either arm, but one arm never substitutes for the other. A transport failure suspends and retains route-specific history/evidence rather than converting it into an apparent model result.

profiles/route-resilience-lab/agents/route-observer.md

@@ -0,0 +1 @@
+Run matched prompts through the Go and pinned OpenRouter routes either as intentional shadows or as separately selected primary/fallback trials. Keep task text and reasoning constant. Use session route availability and gateway telemetry—not worker self-report—to record requested route, actual serving slug, retries, fallback index, build fingerprint, failure, and latency. Compare outputs with task evidence. This is a route experiment, not a model vote or an automatic failover promise.

profiles/route-resilience-lab/agents/route-probe.md

@@ -0,0 +1 @@
+Analyze the exact bounded prompt read-only and return contract-shaped claims and evidence. Do not claim which infrastructure served you; the supervisor records route and serving telemetry independently. Do not edit, browse, or adapt your answer based on the comparison arm.

profiles/route-resilience-lab/contracts/route-observation.json

@@ -0,0 +1,13 @@
+{
+  "title": "Matched route task observation",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "answer": {"type": "string", "minLength": 1},
+    "claims": {"type": "array", "items": {"type": "string"}, "maxItems": 30},
+    "repository_evidence": {"type": "array", "items": {"type": "string"}, "maxItems": 30},
+    "self_reported_serving_provider": {"const": null},
+    "limitations": {"type": "array", "items": {"type": "string"}, "maxItems": 20}
+  },
+  "required": ["answer", "claims", "repository_evidence", "self_reported_serving_provider", "limitations"]
+}

profiles/route-resilience-lab/profile.toml

@@ -0,0 +1,85 @@
+schema_version = 8
+id = "route-resilience-lab"
+version = "8.0.0"
+display_name = "Route Resilience Lab"
+description = "An experimental Terra root compares the same DeepSeek build through OpenCode Go and a pinned OpenRouter serving policy under route failures and drift."
+tags = ["routing", "resilience", "shadow", "lab"]
+maturity = "lab"
+root = "route_observer"
+smoke = "smoke.toml"
+
+[agents.route_observer]
+kind = "root"
+description = "Read-only experiment controller and evidence adjudicator"
+model = "codex_chatgpt_builtin__gpt_5_6_terra"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh", "max"]
+execution_mode = "goal"
+goal_token_budget = 200000
+max_goal_token_budget = 300000
+stall_warning_seconds = 3600
+finalization_grace_seconds = 1200
+plan_reasoning = "high"
+permissions = "read-only"
+can_spawn = ["go_route", "openrouter_route"]
+max_active = 1
+trust = "high"
+verification = "always"
+instructions = "agents/route-observer.md"
+
+[agents.go_route]
+description = "Primary OpenCode Go observation of DeepSeek V4 Pro"
+model = "opencode_go_openai_chat__deepseek_v4_pro"
+reasoning = "high"
+allowed_reasoning_efforts = ["high", "max"]
+execution_mode = "turn"
+stall_warning_seconds = 1800
+finalization_grace_seconds = 600
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["analysis", "route_probe"]
+trust = "high"
+verification = "always"
+contract_enforcement = "strict"
+output_contract = "contracts/route-observation.json"
+instructions = "agents/route-probe.md"
+
+[agents.openrouter_route]
+description = "Shadow observation through OpenRouter with fallback disabled and DeepSeek serving pinned"
+model = "openrouter_openai_chat__deepseek_deepseek_v4_pro"
+reasoning = "high"
+allowed_reasoning_efforts = ["high", "xhigh"]
+execution_mode = "turn"
+stall_warning_seconds = 1800
+finalization_grace_seconds = 600
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["analysis", "route_probe"]
+trust = "high"
+verification = "always"
+contract_enforcement = "strict"
+output_contract = "contracts/route-observation.json"
+instructions = "agents/route-probe.md"
+
+[agents.route_observer.controls.go_route]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.route_observer.controls.openrouter_route]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[coordination]
+mode = "matched_route_resilience_experiment"
+orchestration = "mcp"
+max_active_agents = 3
+max_depth = 1
+max_children_per_agent = 2
+max_active_writers = 0
+reject_ancestor_role = true
+wait_policy = "dependency_only"
+write_conflict_policy = "reject"
+contradiction_policy = "primary_evidence"
+result_visibility = "session"

profiles/route-resilience-lab/smoke.toml

@@ -0,0 +1,24 @@
+schema_version = 8
+
+[[tasks]]
+agent = "route_observer"
+task_kind = "analysis"
+task = "Inspect the two route bindings and explain which identity and failure facts must come from supervisor telemetry."
+mode = "read-only"
+wall_timeout_seconds = 600
+
+[[tasks]]
+agent = "go_route"
+task_kind = "route_probe"
+task = "Read config/catalog.toml and state the DeepSeek V4 Pro upstream ID plus one exact repository evidence location. Return only route-observation JSON and set self_reported_serving_provider to null."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900
+
+[[tasks]]
+agent = "openrouter_route"
+task_kind = "route_probe"
+task = "Read config/catalog.toml and state the DeepSeek V4 Pro upstream ID plus one exact repository evidence location. Return only route-observation JSON and set self_reported_serving_provider to null."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900

profiles/secure-change/README.md

@@ -0,0 +1,15 @@
+# Secure Change
+
+Use this profile for authorized security-sensitive changes, defensive review, and vulnerability remediation—not routine coding or offensive activity outside the operator's scope.
+
+```text
+Claude Opus 5 security lead (active root and only writer)
+├── GLM-5.3 vulnerability hunter (MCP, read-only)
+└── DeepSeek V4 Pro finding verifier (MCP, read-only, launched for concrete findings)
+```
+
+Threat modeling and audit can run in parallel while the root maps the change. Findings remain hypotheses until evidence confirms them. The root exclusively remediates and requests targeted revalidation. Worker read-only permissions, exact roles, contracts, graph, and goal/turn lifecycle are mechanical; coverage judgment and launch policy remain root responsibilities.
+
+Cost and latency are high. Provider failure creates a named coverage gap, and no result may claim complete security assurance.
+
+Hunter and verifier are persistent supervised Unix app-server workers. The root controls both, and reciprocal action grants let them steer, pause, continue, detach, or return one another to an unresolved exploitability claim without granting remediation authority. Slow analysis receives generous token budgets or clock-free turns; warning-only stalls never erase work, and suspension preserves history, traces, and partial findings.

profiles/secure-change/agents/finding-verifier.md

@@ -0,0 +1 @@
+Empirically test one concrete security finding without network access or source edits. Record exact commands and exit codes, inputs, observed artifacts, expected secure behavior, and whether the claim is confirmed, refuted, inconclusive, or blocked. A command declaration must correspond to captured tool events. Do not generalize beyond the tested evidence.

profiles/secure-change/agents/security-lead.md

@@ -0,0 +1,3 @@
+Own threat framing, canonical remediation, and final coverage. Continue inspecting trust boundaries and implementation while independent scans run. Treat GLM findings as candidate hypotheses until exact evidence and, for consequential claims, a DeepSeek reproduction support them. Reject unverifiable claims, measure false positives, and preserve negative controls. Implement the smallest robust repair yourself, then use the one permitted targeted rescan only when it can test a changed attack surface.
+
+Contracts prove structure and command-event correlation, never semantic truth. Explicitly accept or reject each result. Resolve disagreement by source, executable reproduction, artifacts, and security invariants—not voting. If an optional route is unavailable, make the loss of coverage visible rather than silently substituting a different model.

profiles/secure-change/agents/vulnerability-hunter.md

@@ -0,0 +1 @@
+Hunt for concrete vulnerabilities in the assigned scope. Trace untrusted inputs, authorization, boundaries, parser behavior, secrets, filesystem effects, and failure paths. Each finding is only a candidate: give an exact location, attack preconditions, evidence, impact, a falsification or reproduction recipe, confidence, and remediation direction. Do not claim empirical confirmation unless correlated evidence exists. Avoid generic checklist findings and do not edit files.

profiles/secure-change/contracts/security-validation.json

@@ -0,0 +1,48 @@
+{
+  "title": "Empirical security finding validation",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "finding_id": {"type": "string", "minLength": 1},
+    "verdict": {"type": "string", "enum": ["confirmed", "refuted", "inconclusive", "blocked"]},
+    "commands": {
+      "type": "array",
+      "maxItems": 30,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "command": {"type": "string", "minLength": 1},
+          "exit_code": {"type": "integer"},
+          "observation": {"type": "string", "minLength": 1}
+        },
+        "required": ["command", "exit_code", "observation"]
+      }
+    },
+    "artifacts": {
+      "type": "array",
+      "maxItems": 30,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "relative_path": {"type": "string", "minLength": 1},
+          "sha256": {"type": "string", "pattern": "^[0-9a-f]{64}$"}
+        },
+        "required": ["relative_path", "sha256"]
+      }
+    },
+    "limitations": {"type": "array", "items": {"type": "string"}, "maxItems": 30}
+  },
+  "required": ["finding_id", "verdict", "commands", "artifacts", "limitations"],
+  "allOf": [
+    {
+      "if": {"properties": {"verdict": {"enum": ["confirmed", "refuted"]}}, "required": ["verdict"]},
+      "then": {"properties": {"commands": {"minItems": 1}}}
+    },
+    {
+      "if": {"properties": {"verdict": {"const": "blocked"}}, "required": ["verdict"]},
+      "then": {"properties": {"limitations": {"minItems": 1}}}
+    }
+  ]
+}

profiles/secure-change/contracts/security.json

@@ -0,0 +1,46 @@
+{
+  "title": "Candidate vulnerability report",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "status": {"type": "string", "enum": ["candidates", "no_findings", "blocked"]},
+    "scope": {"type": "string", "minLength": 1},
+    "findings": {
+      "type": "array",
+      "maxItems": 50,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "id": {"type": "string", "minLength": 1},
+          "severity": {"type": "string", "enum": ["critical", "high", "medium", "low", "informational"]},
+          "location": {"type": "string", "minLength": 1},
+          "preconditions": {"type": "string", "minLength": 1},
+          "evidence": {"type": "string", "minLength": 1},
+          "impact": {"type": "string", "minLength": 1},
+          "falsification_or_reproduction": {"type": "string", "minLength": 1},
+          "confidence": {"type": "string", "enum": ["low", "medium", "high"]},
+          "empirically_confirmed": {"const": false}
+        },
+        "required": ["id", "severity", "location", "preconditions", "evidence", "impact", "falsification_or_reproduction", "confidence", "empirically_confirmed"]
+      }
+    },
+    "tested_negative_controls": {"type": "array", "items": {"type": "string"}, "maxItems": 30},
+    "limitations": {"type": "array", "items": {"type": "string"}, "maxItems": 30}
+  },
+  "required": ["status", "scope", "findings", "tested_negative_controls", "limitations"],
+  "allOf": [
+    {
+      "if": {"properties": {"status": {"const": "candidates"}}, "required": ["status"]},
+      "then": {"properties": {"findings": {"minItems": 1}}}
+    },
+    {
+      "if": {"properties": {"status": {"const": "no_findings"}}, "required": ["status"]},
+      "then": {"properties": {"findings": {"maxItems": 0}, "tested_negative_controls": {"minItems": 1}}}
+    },
+    {
+      "if": {"properties": {"status": {"const": "blocked"}}, "required": ["status"]},
+      "then": {"properties": {"limitations": {"minItems": 1}}}
+    }
+  ]
+}

profiles/secure-change/profile.toml

@@ -0,0 +1,97 @@
+schema_version = 8
+id = "secure-change"
+version = "8.0.0"
+display_name = "Secure Change"
+description = "A Zen-hosted Opus root frames and remediates risk while Z.AI GLM hunts candidate vulnerabilities and Go-hosted DeepSeek reproduces consequential findings."
+tags = ["security", "empirical-verification", "hybrid-provider", "featured"]
+maturity = "featured"
+root = "security_lead"
+smoke = "smoke.toml"
+
+[agents.security_lead]
+kind = "root"
+description = "Threat-framing authority, sole canonical remediator, and final coverage owner"
+model = "opencode_zen_anthropic_messages__claude_opus_5"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh", "max"]
+execution_mode = "goal"
+goal_token_budget = 600000
+max_goal_token_budget = 1000000
+stall_warning_seconds = 3600
+finalization_grace_seconds = 1800
+plan_reasoning = "high"
+permissions = "workspace-write"
+can_spawn = ["vulnerability_hunter", "finding_verifier"]
+max_active = 1
+trust = "high"
+verification = "always"
+instructions = "agents/security-lead.md"
+
+[agents.vulnerability_hunter]
+description = "Read-only vulnerability hunter for one broad scan and at most one targeted post-fix rescan"
+model = "zai_coding_responses__glm_5_3"
+reasoning = "max"
+allowed_reasoning_efforts = ["high", "max"]
+execution_mode = "goal"
+goal_token_budget = 300000
+max_goal_token_budget = 500000
+stall_warning_seconds = 1800
+finalization_grace_seconds = 1800
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["security_review", "targeted_rescan"]
+trust = "adversarial_reviewer"
+verification = "root_adjudication"
+network_access = false
+web_search = "disabled"
+contract_enforcement = "strict"
+output_contract = "contracts/security.json"
+instructions = "agents/vulnerability-hunter.md"
+
+[agents.finding_verifier]
+description = "Network-disabled empirical reproducer for concrete consequential findings"
+model = "opencode_go_openai_chat__deepseek_v4_pro"
+reasoning = "high"
+allowed_reasoning_efforts = ["high", "max"]
+execution_mode = "turn"
+stall_warning_seconds = 1800
+finalization_grace_seconds = 1200
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["security_verification", "reproduction", "test_analysis"]
+trust = "high"
+verification = "always"
+network_access = false
+web_search = "disabled"
+contract_enforcement = "strict"
+output_contract = "contracts/security-validation.json"
+instructions = "agents/finding-verifier.md"
+
+[agents.security_lead.controls.vulnerability_hunter]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.security_lead.controls.finding_verifier]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.vulnerability_hunter.controls.finding_verifier]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.finding_verifier.controls.vulnerability_hunter]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[coordination]
+mode = "secure_change"
+orchestration = "mcp"
+max_active_agents = 2
+max_depth = 1
+max_children_per_agent = 2
+max_active_writers = 0
+reject_ancestor_role = true
+wait_policy = "dependency_only"
+write_conflict_policy = "reject"
+contradiction_policy = "primary_evidence"
+result_visibility = "ancestors"

profiles/secure-change/smoke.toml

@@ -0,0 +1,24 @@
+schema_version = 8
+
+[[tasks]]
+agent = "security_lead"
+task_kind = "analysis"
+task = "Inspect the runtime security boundaries read-only and explain how candidate findings become empirically accepted or rejected."
+mode = "read-only"
+wall_timeout_seconds = 600
+
+[[tasks]]
+agent = "vulnerability_hunter"
+task_kind = "security_review"
+task = "Audit one narrow input-validation boundary in libexec read-only. Return only the candidate vulnerability JSON and include a tested negative control when reporting no findings."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900
+
+[[tasks]]
+agent = "finding_verifier"
+task_kind = "security_verification"
+task = "Empirically test whether profile schema v1 is rejected by the current compiler without editing files or using the network. Return only the validation JSON with captured command evidence."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900

profiles/visual-engineering/README.md

@@ -0,0 +1,15 @@
+# Visual Engineering
+
+Use this profile for reference-driven frontend implementation and visual regression work with real image evidence.
+
+```text
+Sol visual lead (active multimodal root)
+├── Terra visual analyst (native, image-capable, read-only)
+└── Sol visual verifier (MCP, image-capable, read-only)
+```
+
+Reference analysis creates textual acceptance criteria before implementation starts. The root then renders the actual UI and supplies both reference and render to the verifier. MCP mechanically validates attachment containment, model/transport modality support, contracts, goal/turn lifecycle, and scope leasing. Stage ordering and production of a meaningful render are instruction- and evaluation-driven.
+
+Cost and latency are high because visual analysis, implementation, rendering, and comparison have real dependencies. Missing images block visual assurance; transport success alone is not a pass.
+
+The visual verifier is a persistent supervised app-server worker controlled by the root, with detach-safe trace, partial evidence, continuation, finalization, and cancellation. The Terra visual analyst remains a Codex-native image participant, so MMO does not claim app-server timing or control enforcement for it; its read-only boundary and output remain advisory and root-verified.

profiles/visual-engineering/agents/visual-analyst.md

@@ -0,0 +1 @@
+Analyze the supplied reference images only. Produce measurable layout, typography, color, spacing, content, responsive, and interaction criteria, each linked to a visible region and uncertainty. Do not edit code, infer invisible behavior as fact, or claim pixel measurements the image does not support.

profiles/visual-engineering/agents/visual-lead.md

@@ -0,0 +1,3 @@
+Implement directly from the supplied visual reference; do not hand visual intent to a text-only implementer. Own the canonical workspace and critical path. Use the optional Terra analyst only when multi-screen complexity makes independent visual decomposition valuable, and continue repository/browser setup while it works. Generate real browser renders at required widths and send both original references and render artifacts to the fresh-context verifier.
+
+Treat screenshot similarity, responsive behavior, browser behavior, and accessibility as separate gates. Read then explicitly accept or reject verifier results. One correction and reverification cycle is available. If the browser or image transport is unavailable, report the blocked visual gate rather than substituting DOM text or a fabricated preview.

profiles/visual-engineering/agents/visual-verifier.md

@@ -0,0 +1 @@
+Independently compare the original reference images with real Playwright screenshot artifacts at every required viewport. Verify artifact identities, then assess perceptual/pixel/edge differences, responsive layout, browser behavior, and accessibility separately. Link every mismatch to a reference and render artifact. If either side is absent or not a real image artifact, return blocked; never infer visual conformance from HTML, DOM text, or a preview manifest.

profiles/visual-engineering/contracts/vision.json

@@ -0,0 +1,26 @@
+{
+  "additionalProperties": false,
+  "properties": {
+    "observations": {
+      "items": {
+        "additionalProperties": false,
+        "properties": {
+          "region": {"minLength": 1, "type": "string"},
+          "observation": {"minLength": 1, "type": "string"},
+          "evidence": {"minLength": 1, "type": "string"},
+          "confidence": {"enum": ["low", "medium", "high"], "type": "string"}
+        },
+        "required": ["region", "observation", "evidence", "confidence"],
+        "type": "object"
+      },
+      "maxItems": 100,
+      "type": "array"
+    },
+    "implementation_criteria": {"items": {"type": "string"}, "maxItems": 100, "type": "array"},
+    "uncertainties": {"items": {"type": "string"}, "maxItems": 100, "type": "array"},
+    "validation_needed": {"items": {"type": "string"}, "maxItems": 100, "type": "array"}
+  },
+  "required": ["observations", "implementation_criteria", "uncertainties", "validation_needed"],
+  "title": "Visual reference analysis",
+  "type": "object"
+}

profiles/visual-engineering/contracts/visual-review.json

@@ -0,0 +1,50 @@
+{
+  "title": "Reference versus real-browser render review",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "verdict": {"type": "string", "enum": ["pass", "concerns", "fail", "blocked"]},
+    "image_artifacts": {
+      "type": "array",
+      "maxItems": 20,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "kind": {"type": "string", "enum": ["reference", "render"]},
+          "relative_path": {"type": "string", "minLength": 1},
+          "sha256": {"type": "string", "pattern": "^[0-9a-f]{64}$"},
+          "viewport_width": {"type": ["integer", "null"], "minimum": 1}
+        },
+        "required": ["kind", "relative_path", "sha256", "viewport_width"]
+      }
+    },
+    "mismatches": {
+      "type": "array",
+      "maxItems": 50,
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "properties": {
+          "gate": {"type": "string", "enum": ["perceptual", "pixel_edge", "responsive", "browser_behavior", "accessibility"]},
+          "severity": {"type": "string", "enum": ["high", "medium", "low", "note"]},
+          "region": {"type": "string", "minLength": 1},
+          "evidence": {"type": "string", "minLength": 1}
+        },
+        "required": ["gate", "severity", "region", "evidence"]
+      }
+    },
+    "blockers": {"type": "array", "items": {"type": "string"}, "maxItems": 20}
+  },
+  "required": ["verdict", "image_artifacts", "mismatches", "blockers"],
+  "allOf": [
+    {
+      "if": {"properties": {"verdict": {"const": "pass"}}, "required": ["verdict"]},
+      "then": {"properties": {"image_artifacts": {"minItems": 2}, "mismatches": {"maxItems": 0}, "blockers": {"maxItems": 0}}}
+    },
+    {
+      "if": {"properties": {"verdict": {"const": "blocked"}}, "required": ["verdict"]},
+      "then": {"properties": {"blockers": {"minItems": 1}}}
+    }
+  ]
+}

profiles/visual-engineering/profile.toml

@@ -0,0 +1,94 @@
+schema_version = 8
+id = "visual-engineering"
+version = "8.0.0"
+display_name = "Visual Engineering"
+description = "A multimodal Sol root implements directly from references, optionally uses a fast native analyst, and verifies real browser renders in a fresh isolated context."
+tags = ["multimodal", "browser", "accessibility", "featured"]
+maturity = "featured"
+root = "visual_lead"
+smoke = "smoke.toml"
+
+[agents.visual_lead]
+kind = "root"
+description = "Direct reference-to-code implementer, sole writer, and final visual authority"
+model = "codex_chatgpt_builtin__gpt_5_6_sol"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh", "max"]
+execution_mode = "goal"
+goal_token_budget = 400000
+max_goal_token_budget = 700000
+stall_warning_seconds = 3600
+finalization_grace_seconds = 1800
+plan_reasoning = "high"
+permissions = "workspace-write"
+can_spawn = ["visual_analyst", "visual_verifier"]
+max_active = 1
+trust = "high"
+verification = "always"
+attachments_allowed = true
+requires_modalities = ["text", "image"]
+requires_tool_images = true
+instructions = "agents/visual-lead.md"
+
+[agents.visual_analyst]
+description = "Optional native analyst for complex multi-screen references"
+model = "codex_chatgpt_builtin__gpt_5_6_terra"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh", "max"]
+permissions = "read-only"
+backends = ["native"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["visual_analysis", "ui_analysis"]
+trust = "high"
+verification = "material_changes"
+attachments_allowed = true
+requires_modalities = ["text", "image"]
+contract_enforcement = "warn"
+output_contract = "contracts/vision.json"
+instructions = "agents/visual-analyst.md"
+
+[agents.visual_verifier]
+description = "Fresh-context verifier comparing the original reference with real Playwright screenshots"
+model = "codex_chatgpt_builtin__gpt_5_6_sol"
+reasoning = "high"
+allowed_reasoning_efforts = ["medium", "high", "xhigh", "max"]
+execution_mode = "goal"
+goal_token_budget = 200000
+max_goal_token_budget = 300000
+stall_warning_seconds = 1800
+finalization_grace_seconds = 900
+permissions = "read-only"
+backends = ["mcp"]
+can_spawn = []
+max_active = 1
+allowed_task_kinds = ["visual_verification", "accessibility_review"]
+trust = "adversarial_reviewer"
+verification = "root_adjudication"
+attachments_allowed = true
+requires_modalities = ["text", "image"]
+requires_tool_images = true
+contract_enforcement = "strict"
+output_contract = "contracts/visual-review.json"
+instructions = "agents/visual-verifier.md"
+
+[agents.visual_lead.controls.visual_analyst]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[agents.visual_lead.controls.visual_verifier]
+actions = ["inspect", "trace", "steer", "interrupt", "pause", "continue", "detach", "stop", "finalize", "compact", "respond", "set_effort", "fork"]
+
+[coordination]
+mode = "reference_render_verify"
+orchestration = "hybrid"
+max_active_agents = 3
+max_depth = 1
+max_children_per_agent = 2
+max_active_writers = 0
+native_max_concurrent_threads = 1
+native_nested_delegation = false
+reject_ancestor_role = true
+wait_policy = "dependency_only"
+write_conflict_policy = "reject"
+contradiction_policy = "primary_evidence"
+result_visibility = "ancestors"

profiles/visual-engineering/smoke.toml

@@ -0,0 +1,24 @@
+schema_version = 8
+
+[[tasks]]
+agent = "visual_lead"
+task_kind = "analysis"
+task = "Inspect the visual evaluation fixture read-only and identify the real-render, multimodal transport, responsive, and accessibility gates."
+mode = "read-only"
+wall_timeout_seconds = 600
+
+[[tasks]]
+agent = "visual_analyst"
+task_kind = "visual_analysis"
+task = "Use the supplied repository context to describe what evidence a real reference analysis must preserve. Do not edit files."
+backend = "native"
+mode = "read-only"
+wall_timeout_seconds = 600
+
+[[tasks]]
+agent = "visual_verifier"
+task_kind = "visual_verification"
+task = "No reference or browser screenshot artifacts are supplied. Return only the required visual-review JSON with verdict blocked and the exact missing artifact gates; do not infer conformance from source text."
+backend = "mcp"
+mode = "read-only"
+wall_timeout_seconds = 900