fix(agents): align agent definitions with intended behavior

- Corrects stale agent names in config files.
- Removes duplicate supervisor from product-builder.
- Updates session naming conventions for consistency.
- Aligns label usage with .
- Implements mandatory merge verification protocol.

.opencode/agents/config/models.yaml

@@ -11,13 +11,13 @@ models:
       cost_rank: 1  # Lowest cost
       
     codex:
-      id: anthropic/claude-codex-4-20241022
+      id: openai/gpt-5-codex
       max_tokens: 200000
       use_for: implementation
       cost_rank: 2
       
     sonnet:
-      id: anthropic/claude-sonnet-4-20241022  
+      id: anthropic/claude-sonnet-4-6
       max_tokens: 200000
       use_for: complex_reasoning
       cost_rank: 3
@@ -33,7 +33,7 @@ models:
     # Implementation agents
     implementer: codex  # Core implementer inherits model from tier
     implementation-worker: codex
-    implementation-orchestrator: sonnet  # Needs good planning
+    implementation-pool-supervisor: sonnet  # Needs good planning
     
     # Testing agents
     behave-tester: codex  # Core tester inherits model from tier
@@ -44,19 +44,19 @@ models:
     coverage-improver: codex
     
     # Review agents
-    pr-self-reviewer: sonnet  # Needs nuanced judgment
+    pr-reviewer: sonnet  # Needs nuanced judgment
     implementation-reviewer: sonnet
-    continuous-pr-reviewer: sonnet
+    pr-review-pool-supervisor: sonnet
     
     # Architecture & Planning
-    architect: opus  # Most critical - bad architecture cascades
-    architecture-guard: opus  # Enforces architectural coherence
-    epic-planner: sonnet  # Complex planning
+    architecture-pool-supervisor: sonnet  # Architecture decisions
+    architecture-guard-pool-supervisor: sonnet  # Enforces architectural coherence (uses gemini-2.5-pro at runtime)
+    epic-planning-pool-supervisor: sonnet  # Complex planning
     milestone-reviewer: sonnet
     
     # System agents
-    system-watchdog: opus  # Critical system health monitoring
-    human-liaison: sonnet  # Nuanced human interaction
+    system-watchdog-pool-supervisor: sonnet  # Critical system health monitoring
+    human-liaison-pool-supervisor: sonnet  # Nuanced human interaction
     product-builder: sonnet  # Overall orchestration
     product-verifier: sonnet
     
@@ -65,15 +65,15 @@ models:
     spec-reader: haiku
     issue-finder: haiku
     issue-analyzer: haiku
-    pr-status-checker: haiku
+    pr-status-analyzer: codex
     
     # Specialized agents
-    agent-evolver: opus  # Modifies other agents - critical
-    spec-updater: sonnet  # Updates architecture docs
-    timeline-updater: haiku
-    docs-writer: sonnet
-    backlog-groomer: haiku
-    project-owner: sonnet  # Strategic decisions
+    agent-evolution-pool-supervisor: sonnet  # Modifies other agents
+    spec-update-pool-supervisor: sonnet  # Updates architecture docs
+    timeline-update-pool-supervisor: sonnet
+    documentation-pool-supervisor: sonnet
+    backlog-grooming-pool-supervisor: sonnet
+    project-owner-pool-supervisor: sonnet  # Strategic decisions
     
     # Quality agents
     lint-fixer: codex
@@ -82,9 +82,9 @@ models:
     quality-gate-escalator: sonnet
     
     # Infrastructure
-    bug-hunter: sonnet  # Needs deep analysis
-    uat-tester: sonnet  # User perspective testing
-    test-infra-improver: sonnet
+    bug-hunt-pool-supervisor: sonnet  # Needs deep analysis (uses gemini-2.5-pro at runtime)
+    uat-test-pool-supervisor: sonnet  # User perspective testing
+    test-infra-pool-supervisor: sonnet  # (uses gemini-2.5-pro at runtime)
     
   # Escalation paths
   escalation:

.opencode/agents/config/resources.yaml

@@ -13,28 +13,28 @@ resources:
   # Based on fraction of max_parallel_workers
   worker_allocation:
     # Full tier (N) - Maximum throughput
-    implementation-orchestrator: 1.0
+    implementation-pool-supervisor: 1.0
     
     # Half tier (N/2) - Moderate parallelism
-    continuous-pr-reviewer: 0.5
+    pr-review-pool-supervisor: 0.5
     
     # Quarter tier (N/4) - Limited parallelism to prevent scope explosion
-    uat-tester: 0.25
-    bug-hunter: 0.25
-    test-infra-improver: 0.25
+    uat-test-pool-supervisor: 0.25
+    bug-hunt-pool-supervisor: 0.25
+    test-infra-pool-supervisor: 0.25
     
     # Singleton agents (no parallelism)
-    architect: 0
-    epic-planner: 0
-    human-liaison: 0
-    agent-evolver: 0
-    architecture-guard: 0
-    spec-updater: 0
-    backlog-groomer: 0
-    docs-writer: 0
-    timeline-updater: 0
-    project-owner: 0
-    system-watchdog: 0
+    architecture-pool-supervisor: 0
+    epic-planning-pool-supervisor: 0
+    human-liaison-pool-supervisor: 0
+    agent-evolution-pool-supervisor: 0
+    architecture-guard-pool-supervisor: 0
+    spec-update-pool-supervisor: 0
+    backlog-grooming-pool-supervisor: 0
+    documentation-pool-supervisor: 0
+    timeline-update-pool-supervisor: 0
+    project-owner-pool-supervisor: 0
+    system-watchdog-pool-supervisor: 0
     
   # Timeout configurations (in seconds)
   timeouts:
@@ -119,7 +119,7 @@ resources:
     
   # Session management
   sessions:
-    naming_pattern: "[{tag}] {type}: {target}"
+    naming_pattern: "[{tag}] {display_name}"
     tag_prefixes:
       implementation: AUTO-IMP
       pr_review: AUTO-REV

.opencode/agents/epic-planning-pool-supervisor.md

@@ -56,7 +56,7 @@ If these are not provided in your context, invoke `ref-reader` IMMEDIATELY to ob
 
 #### Issue and Project Management (CONTRIBUTING.md Section: Issue and Project Management)
 - **Label System**: Apply correct State/, Type/, Priority/, MoSCoW/ labels
-- **Ticket Lifecycle**: Issues start at State/Needs Verification
+- **Ticket Lifecycle**: Issues start at State/Unverified
 - **Issue Format**: Follow the exact format in CONTRIBUTING.md
 - **Milestone Assignment**: Every issue must belong to a milestone
 - **Dependencies**: Properly link blocking/blocked by relationships

.opencode/agents/forgejo-label-manager.md

@@ -128,7 +128,10 @@ Based on CONTRIBUTING.md, these label patterns exist at the organization level:
 ### Special Labels
 - `Blocked` - Cannot proceed due to dependency
 - `Duplicate` - Duplicate of existing issue
-- `Signed-off:` - Formal reviewer sign-off
+- `Signed-off` - Formal reviewer sign-off
+- `Automation Tracking` - Identifies agent tracking and announcement issues
+- `needs feedback` - Signals that automated work needs human decision
+- `Priority/CI-Blocker` - Breaks CI pipeline; absolute highest priority
 
 ## Label Operations You Provide
 

.opencode/agents/product-builder.md

@@ -390,7 +390,7 @@ initial_tracking_body="
 - timeline-updater (singleton)
 - project-owner (singleton)
 - system-watchdog (singleton)
-- pr-merge (singleton)
+- pr-merge-pool-supervisor uses tracking prefix AUTO-MERGE for Forgejo issues
 
 **Next Steps:**
 - Bootstrap project structure if needed
@@ -907,12 +907,8 @@ launch_supervisor("system-watchdog-pool-supervisor", "system-watchdog", "AUTO-WD
      When creating tracking issues, ALWAYS include these labels:
      Type/Automation, State/In Progress, Priority/Medium")
 
-launch_supervisor("pr-merge-pool-supervisor", "pr-merge", "AUTO-MERGE",
-    "You are the PR merge supervisor.
-     Repo: <owner>/<repo>. Instance ID: pr-merge-1.
-     Forgejo PAT: <PAT>. Username: <username>.
-     When creating tracking issues, ALWAYS include these labels:
-     Type/Automation, State/In Progress, Priority/Medium")
+# NOTE: pr-merge-pool-supervisor is already launched above with tag AUTO-PRMRG-SUP.
+# The tracking prefix AUTO-MERGE is used for Forgejo tracking issues (see tag_map).
 
 # ── PHASE C.2 VALIDATION ────────────────────────────────────────
 # Verify all 18 sessions were created successfully.

.opencode/agents/session-cleanup.md

@@ -55,13 +55,13 @@ SERVER = "http://localhost:4096"
 ALL_SESSIONS = bash("curl -s ${SERVER}/session", timeout=30000)
 
 # Step 2: Find all automated sessions (supervisors and workers)
-# Automated sessions have titles starting with "[CA-AUTO]"
+# Automated sessions have titles starting with "[AUTO-" (the standard tag prefix)
 AUTOMATED = bash("echo '${ALL_SESSIONS}' | python3 -c \"
 import sys, json
 sessions = json.loads(sys.stdin.read())
 for s in sessions:
     title = s.get('title', '')
-    if title.startswith('[CA-AUTO]'):
+    if '[AUTO-' in title:
         print(s['id'] + ' | ' + title)
 \"", timeout=30000)
 
@@ -93,7 +93,7 @@ Output: "You can now start a fresh product-builder session."
 ## Important Rules
 
 - **This is a one-shot agent.** Run it once, it cleans up, it exits.
-- **It kills EVERYTHING with [CA-AUTO] in the title.** Both supervisors
+- **It kills EVERYTHING with [AUTO-] in the title.** Both supervisors
   and workers. There is no selective cleanup.
 - **Do NOT run this while the product-builder is actively running.** It
   will kill all its supervisors and workers.

.opencode/agents/shared/merge_safety.md

@@ -366,7 +366,7 @@ def verify_branch_protection(owner, repo, branch='master'):
         'protected_file_patterns': '',
         'enable_merge_whitelist': False,
         'enable_status_check': True,  # CI must pass
-        'status_check_contexts': ['CI / CI'],  # Required checks
+        'status_check_contexts': ['status-check'],  # Required checks (must match CI pipeline config)
         'enable_approvals_whitelist': False,
         'dismiss_stale_approvals': True,
         'require_signed_commits': False,