@mrciphersmith/keryx 0.2.72 → 0.2.74

package/src/gdskills/bundled/skills/orchestration/task-implementer/orchestrator-prompt.md

@@ -1,25 +1,30 @@
 # Task Implementer — Orchestrator Prompt
 
-> **Purpose:** Template used by `wave-executor` (dispatched by `job-orchestrator`) to dispatch `task-implementer` as a sub-agent.
-> The wave-executor fills in the placeholders below and sends the result as a Task prompt.
-> The task-implementer executes SKILL.md autonomously, writes the full JSON result to a file, and returns a compact STATUS response.
+> **Purpose:** Template `job-orchestrator` fills in to dispatch `task-implementer` as a sub-agent.
+> The orchestrator fills in the placeholders below and sends the result as a Task prompt.
+> The task-implementer executes SKILL.md autonomously, writes the full JSON result to a file, records it with `keryx job document`, and returns a compact STATUS response.
 
 ## Data Flow
 
 ```
-[job-orchestrator] → dispatches wave-executor per wave (not task-implementer directly)
-                                    ↓
-[wave-executor]    → extracts tasks → fills template → Task(task-implementer) × N (parallel)
+[job-orchestrator] → extracts tasks → fills template → Task(task-implementer) × N (parallel)
                                                                   ↓
 [task-implementer] → executes SKILL.md → writes JSON to .metaproject/jobs/<job-name>/results/<task_id>.json
+                                        → keryx job document <job-name> --type implementation-report
                                                                   ↓
 [Result]           → compact STATUS: DONE response (no inline JSON)
-                                    ↓
-[wave-executor]    → collects STATUS responses → returns compact WAVE_DONE summary
-                                    ↓
-[job-orchestrator] → receives one-line wave summary per wave
+                                                                  ↓
+[job-orchestrator] → collects STATUS responses; reads a result file only for
+                     DONE_WITH_CONCERNS or BLOCKED
 ```
 
+**There is no `wave-executor` agent.** The orchestrator dispatches every task in
+a wave itself, in one turn, and collects the STATUS lines itself — the same
+statement `job-orchestrator/SKILL.md` makes. This document described a
+`wave-executor` as real in four places while the agent it named existed in no
+catalogue and no dispatcher; `src/job/plans.ts` names `task-implementer` as the
+`implement` and `fix` step agents, with nothing in between.
+
 ## Step 1: Extract Task Parameters
 
 From the issue-analyzer JSON output, extract one task object:
@@ -30,7 +35,7 @@ TASK:
   task_name            → from tasks[i].task_name
   task_type            → from tasks[i].task_type
   complexity           → from tasks[i].complexity
-  dependencies         → from tasks[i].dependencies (already satisfied — dispatch in dependency_order)
+  dependencies         → from tasks[i].dependencies (dispatch in dependency_order)
   description          → from tasks[i].description
   target_files         → from tasks[i].target_files
   acceptance_criteria  → from tasks[i].acceptance_criteria
@@ -52,11 +57,24 @@ JOB_CONTEXT (optional):
 
 ## Step 2: Validate
 
+Write the assembled `{ task, workspace, automation }` object to a file and run:
+
+```bash
+keryx skills contracts validate <request.json> --schema task-implementer-input
 ```
-ASSERT task_id is not empty         → otherwise ABORT: "Missing task_id"
-ASSERT task_type in valid types     → otherwise ABORT: "Invalid task_type"
-ASSERT target_files is not empty    → otherwise ABORT: "No target files"
-ASSERT codebase_path exists         → otherwise ABORT: "Codebase path not found"
+
+Non-zero exit means the request is malformed — the command names the field and
+the rule. Do not dispatch a request that does not validate. The schema is
+`input-contract.schema.json` in this folder, registered in
+`src/gdskills/contracts.ts`; it is what refuses an empty `target_files`, a
+`task_type` outside the enum, a `task_id` that is not `task-<n>`, a missing
+`codebase_path`/`branch`/`issue_number`, `skip_confirmation` other than `true`,
+`max_self_fix_attempts` above 3, and any undeclared field.
+
+One check the schema cannot make, because it is a fact about the machine:
+
+```bash
+test -d "<CODEBASE_PATH>"   # otherwise ABORT: "Codebase path not found"
 ```
 
 ## Step 3: Build Sub-Agent Prompt (regular task)
@@ -65,7 +83,7 @@ ASSERT codebase_path exists         → otherwise ABORT: "Codebase path not foun
 You are running the task-implementer skill in AUTONOMOUS MODE.
 DO NOT ask the user any questions. Execute the full workflow end-to-end.
 
-Load the skill: task-implementer (from skills/task-implementer/SKILL.md)
+Load the skill: task-implementer (from skills/gdskills/orchestration/task-implementer/SKILL.md)
 
 ═══════════════════════════════════════════════
   TASK
@@ -116,7 +134,12 @@ CONTEXT_PATH: <JOBS_ROOT>/<JOB_NAME>/ai/context.md
 
 1. Load task-implementer SKILL.md
 2. Execute all 6 phases: RECEIVE → RESEARCH → PLAN → IMPLEMENT → VERIFY → REPORT
-3. Phase 6: Write full JSON result to <JOBS_ROOT>/<JOB_NAME>/results/<task_id>.json
+3. Phase 6: Write full JSON result to <JOBS_ROOT>/<JOB_NAME>/results/<task_id>.json,
+   then validate and record it:
+     keryx skills contracts validate <that file> --schema task-implementer-output
+     keryx job document <JOB_NAME> --type implementation-report --file <that file>
+   Both refuse rather than warn; `job document` refuses outright if the file was
+   never written.
 4. Return a compact STATUS response as your FINAL MESSAGE — NO inline JSON:
 
    STATUS: DONE
@@ -139,7 +162,7 @@ For fix tasks dispatched from the review loop:
 You are running the task-implementer skill in FIX MODE (AUTONOMOUS).
 DO NOT ask the user any questions. Execute the full workflow end-to-end.
 
-Load the skill: task-implementer (from skills/task-implementer/SKILL.md)
+Load the skill: task-implementer (from skills/gdskills/orchestration/task-implementer/SKILL.md)
 
 ═══════════════════════════════════════════════
   FIX TASK
@@ -171,7 +194,10 @@ WORKSPACE: (same as above)
 
 AUTOMATION SETTINGS: (same as above)
 
-EXECUTION: Run to completion, return JSON result.
+EXECUTION INSTRUCTIONS: (same as above — including the Phase 6 validate/record
+pair, and the compact STATUS response as the FINAL MESSAGE with NO inline JSON.
+`parseChildResult` throws on any first line that is not `STATUS: <TOKEN>`, so a
+returned JSON object is not a result the orchestrator can read.)
 ```
 
 ---
@@ -181,16 +207,16 @@ EXECUTION: Run to completion, return JSON result.
 ```javascript
 Task({
   description: "Implement task-1: <TASK_NAME>",
-  subagent_type: "general",
+  subagent_type: "general-purpose",
   prompt: "<generated prompt from template above>"
 })
 ```
 
 ---
 
-## Parsing the Result (wave-executor)
+## Parsing the Result
 
-After receiving the sub-agent STATUS response, the wave-executor must:
+After receiving the sub-agent STATUS response, the orchestrator must:
 
 1. Read the STATUS line: `DONE` | `DONE_WITH_CONCERNS` | `BLOCKED`
 2. Extract compact summary from the response:
@@ -203,4 +229,5 @@ After receiving the sub-agent STATUS response, the wave-executor must:
 4. Decision based on STATUS:
    - `DONE` → continue, collect commit hashes from response
    - `DONE_WITH_CONCERNS` → read result file, log concerns, continue
-   - `BLOCKED` → STOP wave, report to orchestrator with reason
+   - `BLOCKED` → STOP the wave and report the reason; the `implement` step stays
+     open, and `keryx job complete` refuses while it is

package/src/gdskills/bundled/skills/orchestration/task-implementer/output-contract.schema.json

@@ -2,12 +2,12 @@
   "$schema": "https://json-schema.org/draft/2020-12/schema",
   "$id": "https://keryx.local/schemas/task-implementer/output-contract.schema.json",
   "title": "Task Implementer Output Contract",
-  "description": "JSON result returned by task-implementer after executing a task. Replaces the previous Gherkin format.",
+  "description": "JSON result task-implementer writes in Phase 6.1. Registered as contract `task-implementer-output`: validate a result with `keryx skills contracts validate <result.json> --schema task-implementer-output`.",
   "type": "object",
   "additionalProperties": false,
   "required": ["task_id", "task_name", "task_type", "status"],
   "properties": {
-    "task_id": { "type": "string", "pattern": "^task-[0-9]+$" },
+    "task_id": { "type": "string", "pattern": "^(?:task|fix)-[0-9]+$" },
     "task_name": { "type": "string" },
     "task_type": {
       "type": "string",
@@ -43,6 +43,10 @@
       "type": "string",
       "description": "all OR 'partial: <list of unmet criteria>' OR none"
     },
+    "skill_drift": {
+      "type": "string",
+      "description": "Set from SKILL.md Phase 2.0b: 'none' | 'stale: <module>/<skill> — <what diverged>' | 'missing: <module> should have a project-skill'. SKILL.md has required this field since project-skill verification was added; this object is `additionalProperties: false`, so until it was declared here a compliant result was REFUSED by its own contract."
+    },
     "notes": {
       "type": "string",
       "description": "Warnings, blockers, or additional context"

package/src/gdskills/bundled/skills/orchestration/task-implementer/task-request.template.md

@@ -6,7 +6,9 @@
   Заполняется оркестратором ПЕРЕД запуском task-implementer.
   Все обязательные поля (*) должны быть заполнены.
   
-  Валидация: input-contract.schema.json
+  Валидация: keryx skills contracts validate <request.json> --schema task-implementer-input
+             (схема — input-contract.schema.json, зарегистрирована в src/gdskills/contracts.ts;
+              команда возвращает ненулевой код и называет поле и правило)
   Использование: orchestrator-prompt.md читает этот файл и формирует промпт для субагента.
 -->
 
@@ -87,8 +89,8 @@ Components in `src/pipelines/components/` use `observer()` wrapping, props inter
 
 | Field | Value |
 |-------|-------|
-| Original Task ID | |
-| Iteration | |
+| Original Task IDs | |
+| Iteration | `1`..`3` |
 
 ### Review Feedback
 
@@ -100,12 +102,16 @@ Components in `src/pipelines/components/` use `observer()` wrapping, props inter
 
 ## Automation Settings *
 
-| Setting | Value | Description |
-|---------|-------|-------------|
-| skip_confirmation * | `true` | Must be true for autonomous mode |
-| auto_commit | `true` | Commit changes after implementation |
-| verify_lint | `true` | Run ESLint |
-| verify_types | `true` | Run type-check |
-| verify_tests | `true` | Run tests |
-| verify_stories | `false` | Build storybook |
-| max_self_fix_attempts | `3` | Max self-fix attempts |
+<!-- Типы, значения по умолчанию и границы объявлены в input-contract.schema.json
+     (объект `automation`) и проверяются командой validate выше. Здесь — только
+     значения для этого запроса. -->
+
+| Setting | Value |
+|---------|-------|
+| skip_confirmation * | `true` |
+| auto_commit | `true` |
+| verify_lint | `true` |
+| verify_types | `true` |
+| verify_tests | `true` |
+| verify_stories | `false` |
+| max_self_fix_attempts | `3` |

package/src/gdskills/bundled/skills/planning/brainstorm/SKILL.codex.md

@@ -1,6 +1,6 @@
 ---
 name: brainstorm
-description: "Structured brainstorming with multiple thinking perspectives (Pragmatist, Innovator, Critic). Generates ideas, evaluates trade-offs via comparison matrix, produces actionable recommendation. Use for architecture decisions, feature ideation, problem solving, tech choices, naming."
+description: "Use when exploring architecture decisions, tech choices, feature ideas, or any open-ended problem that benefits from multiple perspectives."
 triggers:
   - "/brainstorm"
   - "Brainstorm"

package/src/gdskills/bundled/skills/planning/brainstorm/SKILL.cursor.md

@@ -1,6 +1,6 @@
 ---
 name: brainstorm
-description: "Structured brainstorming with multiple thinking perspectives (Pragmatist, Innovator, Critic). Generates ideas, evaluates trade-offs via comparison matrix, produces actionable recommendation. Use for architecture decisions, feature ideation, problem solving, tech choices, naming."
+description: "Use when exploring architecture decisions, tech choices, feature ideas, or any open-ended problem that benefits from multiple perspectives."
 triggers:
   - "/brainstorm"
   - "Brainstorm"

package/src/gdskills/bundled/skills/planning/consistency-checker/SKILL.codex.md

@@ -1,18 +1,15 @@
 ---
-name: gproject-consistency-checker
-description: "Validates PRD against decisions registry, architecture doc, and best practices constraints. Adversarial quality gate. Use when: dispatched by gproject-orchestrator Phase 5."
-triggers:
-  - "gproject-consistency-checker: validate PRD"
-  - "Dispatched by gproject-orchestrator Phase 5"
+name: consistency-checker
+description: >
+  Validates PRD/Implementation Plan against decisions registry, architecture doc,
+  and best practices constraints. Catches contradictions, gaps, and violations.
+  Use when: dispatched by gproject-orchestrator Phase 5.
+  NOT for: direct user invocation.
 metadata:
-  author: "MrCipherSmith"
-  version: "1.0.0"
-  category: "planning"
+  version: 1.0.0
   compatible_harnesses: "cursor,codex,zed,opencode"
-license: "MIT"
 ---
 
-
 # gproject-consistency-checker
 
 ## Purpose
@@ -31,3 +28,165 @@ This agent is adversarial — its job is to find problems, not to approve.
 | 4 | Classify severity: CRITICAL (blocks approval) / WARNING (should fix) / INFO (suggestion) |
 | 5 | NEVER fix violations yourself — only report them |
 | 6 | A clean report still lists what was checked (audit trail) |
+
+---
+
+## Input Contract
+
+```yaml
+task: "Validate PRD consistency against all project decisions and constraints"
+input_artifacts:
+  - .metaproject/jobs/<job>/artifacts/prd.md
+  - .metaproject/jobs/<job>/artifacts/problem-statement.md
+  - .metaproject/jobs/<job>/artifacts/stack-decision.md
+  - .metaproject/jobs/<job>/artifacts/architecture.md
+  - .metaproject/jobs/<job>/artifacts/tech-bestpractices.md
+  - .metaproject/jobs/<job>/decisions.md
+```
+
+## Output Contract
+
+```yaml
+status: "DONE" | "DONE_WITH_CONCERNS"
+summary: "<3-5 sentences: checks run, violations found, overall verdict>"
+concerns: ["<CRITICAL violations if any>"]
+artifact_path: ".metaproject/jobs/<job>/artifacts/consistency-report.md"
+# Note: no new_decisions — this agent doesn't make decisions
+```
+
+Status logic:
+- `DONE` = zero CRITICAL violations (WARNINGs acceptable)
+- `DONE_WITH_CONCERNS` = one or more CRITICAL violations found
+
+---
+
+## Workflow
+
+### Step 1: Goals ↔ User Stories Check
+
+For each goal in problem-statement.md:
+- [ ] At least one user story addresses this goal
+- [ ] User story acceptance criteria map to goal's success metric
+
+For each user story in PRD:
+- [ ] Traces back to a defined goal (not an invented one)
+- [ ] Does NOT implement a non-goal
+
+**Violation types:**
+- CRITICAL: Goal with no user story coverage
+- CRITICAL: User story implementing a non-goal
+- WARNING: User story with weak traceability to goal
+
+### Step 2: Stack ↔ PRD Check
+
+For each technology in stack-decision.md:
+- [ ] PRD technical references use this technology (not something else)
+- [ ] No undeclared technologies appear in PRD
+
+**Violation types:**
+- CRITICAL: PRD references technology not in stack-decision.md
+- CRITICAL: Key stack technology has no presence in PRD
+- WARNING: Stack technology mentioned but underspecified in PRD
+
+### Step 3: Architecture ↔ PRD Check
+
+For each architecture decision in architecture.md:
+- [ ] PRD respects layer boundaries
+- [ ] PRD respects dependency direction
+- [ ] PRD technical notes align with defined patterns
+
+**Violation types:**
+- CRITICAL: PRD proposes pattern contradicting architecture.md
+- WARNING: PRD section doesn't reference applicable architecture decision
+
+### Step 4: Best Practices ↔ PRD Check
+
+For each MUST constraint in tech-bestpractices.md:
+- [ ] PRD does not violate this constraint
+- [ ] Relevant user stories reference this constraint
+
+For each MUST NOT constraint:
+- [ ] PRD does not propose the antipattern
+
+**Violation types:**
+- CRITICAL: PRD violates a MUST constraint
+- CRITICAL: PRD proposes a MUST NOT antipattern
+- WARNING: MUST constraint not referenced by any user story
+
+### Step 5: Internal Consistency Check
+
+Within the PRD itself:
+- [ ] No two user stories contradict each other
+- [ ] Priorities are consistent (P0 depends on P0, not P1 depending on P2)
+- [ ] Data model matches API surface
+- [ ] Security requirements cover all user-facing features
+- [ ] Testing requirements cover all P0 user stories
+
+**Violation types:**
+- CRITICAL: Contradicting user stories
+- CRITICAL: P0 feature with no test coverage requirement
+- WARNING: Priority inconsistency in dependencies
+
+### Step 6: Completeness Check
+
+- [ ] All sections of PRD template are present
+- [ ] Executive summary matches actual content
+- [ ] Traceability matrix is complete
+- [ ] Open questions are flagged (not hidden)
+
+**Violation types:**
+- WARNING: Missing PRD section
+- INFO: Traceability matrix has gaps
+
+### Step 7: Write Consistency Report
+
+Write `artifacts/consistency-report.md`:
+
+```markdown
+# Consistency Report: <Project/Task Name>
+
+## Verdict: PASS | PASS_WITH_WARNINGS | FAIL
+
+## Summary
+- Checks executed: <count>
+- CRITICAL violations: <count>
+- WARNING violations: <count>
+- INFO items: <count>
+
+## Violations
+
+### CRITICAL
+#### V-001: <Title>
+- **Check**: <which check caught this>
+- **Found in PRD**: <section/user story reference>
+- **Conflicts with**: <decision ID / constraint ID / architecture section>
+- **Details**: <specific description>
+- **Suggested fix**: <what should change>
+- **Fix target**: <which phase/artifact needs to be updated>
+
+### WARNINGS
+#### W-001: <Title>
+...
+
+### INFO
+#### I-001: <Title>
+...
+
+## Audit Trail
+| Check Category | Items Checked | Passed | Failed |
+|---------------|--------------|--------|--------|
+| Goals ↔ User Stories | <N> | <N> | <N> |
+| Stack ↔ PRD | <N> | <N> | <N> |
+| Architecture ↔ PRD | <N> | <N> | <N> |
+| Best Practices ↔ PRD | <N> | <N> | <N> |
+| Internal Consistency | <N> | <N> | <N> |
+| Completeness | <N> | <N> | <N> |
+
+## Rollback Recommendations
+<If CRITICAL violations exist, suggest which phase to rollback to>
+- V-001 → requires rollback to Phase <N> (<reason>)
+```
+
+### Step 8: Return Summary
+
+Compact summary: verdict, violation counts, rollback recommendation if applicable.

package/src/gdskills/bundled/skills/planning/consistency-checker/SKILL.cursor.md

@@ -1,18 +1,15 @@
 ---
-name: gproject-consistency-checker
-description: "Validates PRD against decisions registry, architecture doc, and best practices constraints. Adversarial quality gate. Use when: dispatched by gproject-orchestrator Phase 5."
-triggers:
-  - "gproject-consistency-checker: validate PRD"
-  - "Dispatched by gproject-orchestrator Phase 5"
+name: consistency-checker
+description: >
+  Validates PRD/Implementation Plan against decisions registry, architecture doc,
+  and best practices constraints. Catches contradictions, gaps, and violations.
+  Use when: dispatched by gproject-orchestrator Phase 5.
+  NOT for: direct user invocation.
 metadata:
-  author: "MrCipherSmith"
-  version: "1.0.0"
-  category: "planning"
+  version: 1.0.0
   compatible_harnesses: "cursor,codex,zed,opencode"
-license: "MIT"
 ---
 
-
 # gproject-consistency-checker
 
 ## Purpose
@@ -31,3 +28,165 @@ This agent is adversarial — its job is to find problems, not to approve.
 | 4 | Classify severity: CRITICAL (blocks approval) / WARNING (should fix) / INFO (suggestion) |
 | 5 | NEVER fix violations yourself — only report them |
 | 6 | A clean report still lists what was checked (audit trail) |
+
+---
+
+## Input Contract
+
+```yaml
+task: "Validate PRD consistency against all project decisions and constraints"
+input_artifacts:
+  - .metaproject/jobs/<job>/artifacts/prd.md
+  - .metaproject/jobs/<job>/artifacts/problem-statement.md
+  - .metaproject/jobs/<job>/artifacts/stack-decision.md
+  - .metaproject/jobs/<job>/artifacts/architecture.md
+  - .metaproject/jobs/<job>/artifacts/tech-bestpractices.md
+  - .metaproject/jobs/<job>/decisions.md
+```
+
+## Output Contract
+
+```yaml
+status: "DONE" | "DONE_WITH_CONCERNS"
+summary: "<3-5 sentences: checks run, violations found, overall verdict>"
+concerns: ["<CRITICAL violations if any>"]
+artifact_path: ".metaproject/jobs/<job>/artifacts/consistency-report.md"
+# Note: no new_decisions — this agent doesn't make decisions
+```
+
+Status logic:
+- `DONE` = zero CRITICAL violations (WARNINGs acceptable)
+- `DONE_WITH_CONCERNS` = one or more CRITICAL violations found
+
+---
+
+## Workflow
+
+### Step 1: Goals ↔ User Stories Check
+
+For each goal in problem-statement.md:
+- [ ] At least one user story addresses this goal
+- [ ] User story acceptance criteria map to goal's success metric
+
+For each user story in PRD:
+- [ ] Traces back to a defined goal (not an invented one)
+- [ ] Does NOT implement a non-goal
+
+**Violation types:**
+- CRITICAL: Goal with no user story coverage
+- CRITICAL: User story implementing a non-goal
+- WARNING: User story with weak traceability to goal
+
+### Step 2: Stack ↔ PRD Check
+
+For each technology in stack-decision.md:
+- [ ] PRD technical references use this technology (not something else)
+- [ ] No undeclared technologies appear in PRD
+
+**Violation types:**
+- CRITICAL: PRD references technology not in stack-decision.md
+- CRITICAL: Key stack technology has no presence in PRD
+- WARNING: Stack technology mentioned but underspecified in PRD
+
+### Step 3: Architecture ↔ PRD Check
+
+For each architecture decision in architecture.md:
+- [ ] PRD respects layer boundaries
+- [ ] PRD respects dependency direction
+- [ ] PRD technical notes align with defined patterns
+
+**Violation types:**
+- CRITICAL: PRD proposes pattern contradicting architecture.md
+- WARNING: PRD section doesn't reference applicable architecture decision
+
+### Step 4: Best Practices ↔ PRD Check
+
+For each MUST constraint in tech-bestpractices.md:
+- [ ] PRD does not violate this constraint
+- [ ] Relevant user stories reference this constraint
+
+For each MUST NOT constraint:
+- [ ] PRD does not propose the antipattern
+
+**Violation types:**
+- CRITICAL: PRD violates a MUST constraint
+- CRITICAL: PRD proposes a MUST NOT antipattern
+- WARNING: MUST constraint not referenced by any user story
+
+### Step 5: Internal Consistency Check
+
+Within the PRD itself:
+- [ ] No two user stories contradict each other
+- [ ] Priorities are consistent (P0 depends on P0, not P1 depending on P2)
+- [ ] Data model matches API surface
+- [ ] Security requirements cover all user-facing features
+- [ ] Testing requirements cover all P0 user stories
+
+**Violation types:**
+- CRITICAL: Contradicting user stories
+- CRITICAL: P0 feature with no test coverage requirement
+- WARNING: Priority inconsistency in dependencies
+
+### Step 6: Completeness Check
+
+- [ ] All sections of PRD template are present
+- [ ] Executive summary matches actual content
+- [ ] Traceability matrix is complete
+- [ ] Open questions are flagged (not hidden)
+
+**Violation types:**
+- WARNING: Missing PRD section
+- INFO: Traceability matrix has gaps
+
+### Step 7: Write Consistency Report
+
+Write `artifacts/consistency-report.md`:
+
+```markdown
+# Consistency Report: <Project/Task Name>
+
+## Verdict: PASS | PASS_WITH_WARNINGS | FAIL
+
+## Summary
+- Checks executed: <count>
+- CRITICAL violations: <count>
+- WARNING violations: <count>
+- INFO items: <count>
+
+## Violations
+
+### CRITICAL
+#### V-001: <Title>
+- **Check**: <which check caught this>
+- **Found in PRD**: <section/user story reference>
+- **Conflicts with**: <decision ID / constraint ID / architecture section>
+- **Details**: <specific description>
+- **Suggested fix**: <what should change>
+- **Fix target**: <which phase/artifact needs to be updated>
+
+### WARNINGS
+#### W-001: <Title>
+...
+
+### INFO
+#### I-001: <Title>
+...
+
+## Audit Trail
+| Check Category | Items Checked | Passed | Failed |
+|---------------|--------------|--------|--------|
+| Goals ↔ User Stories | <N> | <N> | <N> |
+| Stack ↔ PRD | <N> | <N> | <N> |
+| Architecture ↔ PRD | <N> | <N> | <N> |
+| Best Practices ↔ PRD | <N> | <N> | <N> |
+| Internal Consistency | <N> | <N> | <N> |
+| Completeness | <N> | <N> | <N> |
+
+## Rollback Recommendations
+<If CRITICAL violations exist, suggest which phase to rollback to>
+- V-001 → requires rollback to Phase <N> (<reason>)
+```
+
+### Step 8: Return Summary
+
+Compact summary: verdict, violation counts, rollback recommendation if applicable.

package/src/gdskills/bundled/skills/planning/consistency-checker/SKILL.md

@@ -1,5 +1,5 @@
 ---
-name: gproject-consistency-checker
+name: consistency-checker
 description: >
   Validates PRD/Implementation Plan against decisions registry, architecture doc,
   and best practices constraints. Catches contradictions, gaps, and violations.

package/src/gdskills/bundled/skills/planning/interview/SKILL.codex.md

@@ -1,6 +1,6 @@
 ---
 name: interview
-description: "Critical requirements interviewer: asks targeted clarifying questions one-by-one before expensive operations. Identifies uncertainty zones, provides answer options, adapts follow-ups based on responses. Can trigger mini-brainstorm on ambiguous points. Use before implementation, design, or migration to gather precise context."
+description: "Use before implementation, design, or migration when requirements are unclear and targeted clarifying questions are needed to gather precise context."
 triggers:
   - "/interview"
   - "Interview"
@@ -16,6 +16,12 @@ metadata:
 license: "MIT"
 ---
 
+<SUBAGENT-STOP>
+If you were dispatched as a subagent to execute a specific task, skip this skill entirely.
+This skill is for orchestrators and interactive session-level routing only.
+Proceed directly with your assigned task.
+</SUBAGENT-STOP>
+
 # Interview
 
 Critical requirements interviewer that asks targeted clarifying questions before expensive operations begin. The goal is to eliminate ambiguity and gather precise context so downstream skills (job-orchestrator, feature-dev, prd-creator) produce accurate results.