maestro-flow 0.4.5 → 0.4.7

package/.codex/skills/team-lifecycle-v4/SKILL.md

@@ -1,191 +1,445 @@
----
-name: team-lifecycle-v4
-description: Full lifecycle team -- plan, develop, test, review
-allowed-tools: spawn_agent(*), wait_agent(*), send_message(*), followup_task(*), close_agent(*), list_agents(*), report_agent_job_result(*), Read(*), Write(*), Edit(*), Bash(*), Glob(*), Grep(*), request_user_input(*), mcp__maestro-tools__team_msg(*)
----
-
-<purpose>
-Orchestrate multi-agent software development: specification -> planning -> implementation -> testing -> review.
-
-```
-Skill(skill="team-lifecycle-v4", args="task description")
-                    |
-         SKILL.md (this file) = Router
-                    |
-     +--------------+--------------+
-     |                             |
-  no --role flag              --role <name>
-     |                             |
-  Coordinator                  Worker
-  roles/coordinator/role.md    roles/<name>/role.md
-     |
-     +-- analyze -> dispatch -> spawn -> wait -> collect
-                                 |
-                    +--------+---+--------+
-                    v        v            v
-            spawn_agent    ...     spawn_agent
-          (team_worker)         (team_supervisor)
-              per-task             resident agent
-              lifecycle            followup_task-driven
-                    |                     |
-                    +-- wait_agent --------+
-                              |
-                         collect results
-```
-
-</purpose>
-
-<context>
-### Role Registry
-
-| Role | Path | Prefix | Inner Loop |
-|------|------|--------|------------|
-| coordinator | [roles/coordinator/role.md](roles/coordinator/role.md) | -- | -- |
-| analyst | [roles/analyst/role.md](roles/analyst/role.md) | RESEARCH-* | false |
-| writer | [roles/writer/role.md](roles/writer/role.md) | DRAFT-* | true |
-| planner | [roles/planner/role.md](roles/planner/role.md) | PLAN-* | true |
-| executor | [roles/executor/role.md](roles/executor/role.md) | IMPL-* | true |
-| tester | [roles/tester/role.md](roles/tester/role.md) | TEST-* | false |
-| reviewer | [roles/reviewer/role.md](roles/reviewer/role.md) | REVIEW-*, QUALITY-*, IMPROVE-* | false |
-| supervisor | [roles/supervisor/role.md](roles/supervisor/role.md) | CHECKPOINT-* | false |
-
-### Role Router
-
-Parse `$ARGUMENTS`:
-- Has `--role <name>` -> Read `roles/<name>/role.md`, execute Phase 2-4
-- No `--role` -> `roles/coordinator/role.md`, execute entry router
-
-### Delegation Lock
-
-**Coordinator is a PURE ORCHESTRATOR. It coordinates, it does NOT do.**
-
-Before calling ANY tool, apply this check:
-
-| Tool Call | Verdict | Reason |
-|-----------|---------|--------|
-| `spawn_agent`, `wait_agent`, `close_agent`, `send_message`, `followup_task` | ALLOWED | Orchestration |
-| `list_agents` | ALLOWED | Agent health check |
-| `request_user_input` | ALLOWED | User interaction |
-| `mcp__maestro-tools__team_msg` | ALLOWED | Message bus |
-| `Read/Write` on `.workflow/.team/` files | ALLOWED | Session state |
-| `Read` on `roles/`, `commands/`, `specs/`, `templates/` | ALLOWED | Loading own instructions |
-| `Read/Grep/Glob` on project source code | BLOCKED | Delegate to worker |
-| `Edit` on any file outside `.workflow/` | BLOCKED | Delegate to worker |
-| `Bash("maestro delegate ...")` | BLOCKED | Only workers call CLI |
-| `Bash` running build/test/lint commands | BLOCKED | Delegate to worker |
-
-**If a tool call is BLOCKED**: STOP. Create a task, spawn a worker.
-
-**No exceptions for "simple" tasks.** Even a single-file read-and-report MUST go through spawn_agent.
-
-### Shared Constants
-
-- **Session prefix**: `TLV4`
-- **Session path**: `.workflow/.team/TLV4-<slug>-<date>/`
-- **State file**: `<session>/tasks.json`
-- **Discovery files**: `<session>/discoveries/{task_id}.json`
-- **CLI tools**: `maestro delegate --mode analysis` (read-only), `maestro delegate --mode write` (modifications)
-
-### Worker Spawn Template
-
-Spawn via `team-worker` agent. Message includes: role, role_spec path (`<skill_root>/roles/<role>/role.md`), session folder/id, requirement, inner_loop flag, task context (id, title, description, pipeline_phase), upstream context. Worker executes: Phase 1 (discovery) -> role Phase 2-4 -> Phase 5 (report).
-
-### Supervisor Spawn Template
-
-Supervisor is a **resident agent** (`team_supervisor`), independent from team_worker.
-
-- **Spawn** (Phase 2, once): Load role-spec from `roles/supervisor/role.md`, init baseline context, go idle
-- **Wake** (per CHECKPOINT): `followup_task` with checkpoint task_id, upstream scope, progress. `wait_agent` (30 min)
-- **Shutdown**: `close_agent({ target: "supervisor" })` at pipeline end
-
-### Model Selection Guide
-
-| Role | reasoning_effort |
-|------|-------------------|
-| Analyst (RESEARCH-*) | medium |
-| Writer (DRAFT-*) | high |
-| Planner (PLAN-*) | high |
-| Executor (IMPL-*) | high |
-| Tester (TEST-*) | high |
-| Reviewer (REVIEW-*, QUALITY-*, IMPROVE-*) | high |
-| Supervisor (CHECKPOINT-*) | medium |
-
-Override via `model`/`reasoning_effort` params in spawn_agent for cost optimization.
-
-### v4 Agent Coordination
-
-**Message Semantics**: `send_message` for supplementary info to workers. `followup_task` to wake supervisor for checkpoints. `list_agents` for health checks.
-
-**CRITICAL**: Supervisor is a **resident agent** woken via `followup_task`, NOT `send_message`. Regular workers are one-shot; supervisor persists across checkpoints.
-
-**Agent Health Check**: Reconcile `tasks.json.active_agents` with `list_agents({})`. Reset orphaned tasks to pending. If supervisor missing but CHECKPOINT tasks pending, respawn.
-
-**Named Targeting**: `send_message({ target: "IMPL-001" })`, `followup_task({ target: "supervisor" })`, `close_agent({ target: ... })`.
-
-### User Commands
-
-| Command | Action |
-|---------|--------|
-| `check` / `status` | View execution status graph |
-| `resume` / `continue` | Advance to next step |
-| `revise <TASK-ID> [feedback]` | Revise specific task |
-| `feedback <text>` | Inject feedback for revision |
-| `recheck` | Re-run quality check |
-| `improve [dimension]` | Auto-improve weakest dimension |
-
-### Completion Action
-
-Present choice via `request_user_input`: **Archive & Clean** (recommended), **Keep Active**, **Export Results**.
-
-### Specs Reference
-
-- [specs/pipelines.md](specs/pipelines.md) -- Pipeline definitions and task registry
-- [specs/quality-gates.md](specs/quality-gates.md) -- Quality gate criteria and scoring
-- [specs/knowledge-transfer.md](specs/knowledge-transfer.md) -- Artifact and state transfer protocols
-
-### Session Directory
-
-```
-.workflow/.team/TLV4-<slug>-<date>/
-+-- tasks.json                  # Task state (JSON)
-+-- discoveries/                # Per-task findings ({task_id}.json)
-+-- spec/                       # Spec phase outputs
-+-- plan/                       # Implementation plan
-+-- artifacts/                  # All deliverables
-+-- wisdom/                     # Cross-task knowledge
-+-- explorations/               # Shared explore cache
-+-- discussions/                # Discuss round records
-```
-</context>
-
-<execution>
-
-### Wave Execution Engine
-
-Per wave: load state from `tasks.json` -> skip tasks with failed deps -> build upstream context from `discoveries/{id}.json` -> separate regular vs CHECKPOINT tasks -> spawn regular workers (`wait_agent` 30 min, timeout: STATUS_CHECK 3 min -> FINALIZE 3 min -> close) -> collect results to `discoveries/`, update `tasks.json` -> execute checkpoints via `followup_task` to supervisor -> handle `block` verdict (prompt: Override/Revise/Abort) -> persist state.
-</execution>
-
-<error_codes>
-
-| Scenario | Resolution |
-|----------|------------|
-| Unknown command | Error with available command list |
-| Role not found | Error with role registry |
-| CLI tool fails | Worker fallback to direct implementation |
-| Supervisor crash | Respawn with `recovery: true`, auto-rebuilds from existing reports |
-| Supervisor not ready for CHECKPOINT | Spawn/respawn supervisor, wait for ready, then wake |
-| Completion action fails | Default to Keep Active |
-| Worker timeout | Mark task as failed, continue wave |
-| Discovery file missing | Mark task as failed with "No discovery file produced" |
-</error_codes>
-
-<success_criteria>
-- [ ] Role router correctly dispatches to coordinator or worker based on --role flag
-- [ ] Coordinator spawns workers with correct role-spec paths and task context
-- [ ] Supervisor spawned once and woken via followup_task for checkpoints
-- [ ] Wave execution engine processes tasks in dependency order
-- [ ] Checkpoint verdicts respected (block prompts user, pass continues)
-- [ ] Session state persisted in tasks.json after each wave
-- [ ] Agent health reconciled on resume (orphaned tasks reset to pending)
-- [ ] Completion action presented and handled correctly
-</success_criteria>
+---
+name: team-lifecycle-v4
+description: Full lifecycle team -- plan, develop, test, review
+argument-hint: "[task description] [-y|--yes] [-c|--concurrency N] [--continue] [--pipeline spec-only|impl-only|full-lifecycle]"
+allowed-tools: spawn_agents_on_csv, Read, Write, Edit, Bash, Glob, Grep, request_user_input
+---
+
+<purpose>
+Wave-based lifecycle orchestration via `spawn_agents_on_csv`. Fixed roles with pre-defined pipelines: specification → planning → implementation → testing → review.
+
+**Core workflow**: Select Pipeline → Build CSV from Pipeline Definition → Wave-by-Wave Execution → Aggregate Results
+
+```
++-------------------------------------------------------------------+
+|                LIFECYCLE V4 CSV WAVE WORKFLOW                       |
++-------------------------------------------------------------------+
+|                                                                     |
+|  Phase 1: Pipeline Selection + CSV Generation                      |
+|     +-- Detect pipeline from keywords or --pipeline flag           |
+|     +-- Load pipeline definition from specs/pipelines.md           |
+|     +-- Build tasks.csv with wave assignments                      |
+|     +-- User validates (skip if -y)                                |
+|                                                                     |
+|  Phase 2: Wave Execution Engine                                    |
+|     +-- For each wave N (sequential):                              |
+|     |   +-- Build prev_context from upstream findings              |
+|     |   +-- spawn_agents_on_csv(wave-N.csv)                        |
+|     |   +-- Each agent reads its role.md for domain logic          |
+|     |   +-- Merge results → master tasks.csv                       |
+|     |   +-- CHECKPOINT waves: supervisor verifies quality          |
+|     |   +-- User approval at checkpoints (skip if -y)              |
+|     +-- discoveries.ndjson shared across all waves                 |
+|                                                                     |
+|  Phase 3: Results Aggregation                                      |
+|     +-- Export results.csv                                         |
+|     +-- Generate context.md                                        |
+|     +-- Display summary with deliverables                          |
+|                                                                     |
++-------------------------------------------------------------------+
+```
+
+</purpose>
+
+<context>
+```bash
+$team-lifecycle-v4 "build user authentication system"
+$team-lifecycle-v4 -y --pipeline spec-only "design payment module"
+$team-lifecycle-v4 --pipeline impl-only "implement auth from existing spec"
+$team-lifecycle-v4 --continue "20260518-tlv4-auth-system"
+```
+
+**Flags**:
+- `-y, --yes`: Skip confirmations and checkpoint approvals
+- `-c, --concurrency N`: Max concurrent agents per wave (default: 3)
+- `--continue`: Resume existing session
+- `--pipeline`: Force pipeline (spec-only, impl-only, full-lifecycle)
+- `--no-supervision`: Skip CHECKPOINT tasks (opt-out from supervisor quality gates)
+
+### Role Registry (Fixed)
+
+| Role | Path | Prefix |
+|------|------|--------|
+| analyst | [roles/analyst/role.md](roles/analyst/role.md) | RESEARCH-* |
+| writer | [roles/writer/role.md](roles/writer/role.md) | DRAFT-* |
+| planner | [roles/planner/role.md](roles/planner/role.md) | PLAN-* |
+| executor | [roles/executor/role.md](roles/executor/role.md) | IMPL-* |
+| tester | [roles/tester/role.md](roles/tester/role.md) | TEST-* |
+| reviewer | [roles/reviewer/role.md](roles/reviewer/role.md) | REVIEW-*, QUALITY-* |
+| supervisor | [roles/supervisor/role.md](roles/supervisor/role.md) | CHECKPOINT-* |
+
+### Session Structure
+
+```
+.workflow/.csv-wave/{YYYYMMDD}-tlv4-{slug}/
++-- tasks.csv              (master state)
++-- results.csv            (final export)
++-- discoveries.ndjson     (cross-wave shared)
++-- context.md             (human-readable report)
++-- spec/                  (spec phase outputs)
++-- plan/                  (implementation plan)
++-- artifacts/             (all deliverables)
++-- wave-{N}.csv           (temporary)
++-- wave-{N}-results.csv   (temporary)
+```
+
+**Output**: tasks.csv, results.csv, discoveries.ndjson, context.md, spec/, plan/, artifacts/
+
+### Specs Reference
+
+| Spec | Purpose |
+|------|---------|
+| [specs/pipelines.md](specs/pipelines.md) | Pipeline definitions, task registry, wave assignments |
+| [specs/quality-gates.md](specs/quality-gates.md) | Quality thresholds and scoring dimensions |
+
+### Templates Reference
+
+| Template | Used By |
+|----------|---------|
+| [templates/product-brief.md](templates/product-brief.md) | writer (DRAFT-001) |
+| [templates/requirements.md](templates/requirements.md) | writer (DRAFT-002) |
+| [templates/architecture.md](templates/architecture.md) | writer (DRAFT-003) |
+| [templates/epics.md](templates/epics.md) | writer (DRAFT-004) |
+</context>
+
+<csv_schema>
+
+### tasks.csv (Master State)
+
+```csv
+id,title,description,role,pipeline_phase,deps,context_from,wave
+"RESEARCH-001","Domain research","PURPOSE: Research domain, competitors, constraints | Success: Discovery context packaged\nTASK:\n  - Analyze problem space\n  - Explore codebase patterns\n  - Package context\nEXPECTED: spec/discovery-context.json\nCONSTRAINTS: Read-only","analyst","research","","","1"
+"DRAFT-001","Product brief","PURPOSE: Create product brief | Success: Self-validated document\nTASK:\n  - Apply templates/product-brief.md\n  - Define vision, problem, users, goals\n  - Self-validate against quality-gates.md §3.1\nEXPECTED: spec/product-brief.md\nCONSTRAINTS: Follow template","writer","product-brief","RESEARCH-001","RESEARCH-001","2"
+```
+
+**Input columns** (present in initial tasks.csv and wave-N.csv):
+
+| Column | Description |
+|--------|-------------|
+| `id` | Task ID: `{PREFIX}-{NNN}` |
+| `title` | Short task title |
+| `description` | Task instructions (PURPOSE/TASK/EXPECTED/CONSTRAINTS) |
+| `role` | Fixed role name from registry |
+| `pipeline_phase` | Phase identifier (research, product-brief, requirements, etc.) |
+| `deps` | Semicolon-separated dependency task IDs |
+| `context_from` | Semicolon-separated task IDs for context |
+| `wave` | Wave number |
+
+**Lifecycle columns** (initialized with defaults, updated during execution):
+
+| Column | Initial Value | Description |
+|--------|--------------|-------------|
+| `status` | `pending` | Task lifecycle: pending → completed/failed/blocked/skipped |
+| `findings` | `""` | Populated from output_schema merge |
+| `files_modified` | `""` | Populated from output_schema merge |
+| `quality_score` | `""` | Populated from output_schema merge |
+| `error` | `""` | Populated from output_schema merge |
+
+**Dynamic column** (added to wave-N.csv only, not in initial tasks.csv):
+
+| Column | Description |
+|--------|-------------|
+| `prev_context` | Concatenated findings from context_from tasks, format: `--- TASK-ID: {id} ---\n{findings}` |
+
+**Output columns** (via `output_schema` only, NOT in any CSV input):
+
+| Column | Description |
+|--------|-------------|
+| `result_status` | completed / failed / blocked (maps to master `status`) |
+| `findings` | Key findings (max 500 chars) |
+| `files_modified` | Semicolon-separated paths |
+| `quality_score` | Numeric quality score (0-100) |
+| `error` | Error message if failed |
+
+**Column separation rule**: Input/lifecycle columns and output_schema MUST NOT share names. `result_status` → master `status` during merge.
+
+### Pipeline Wave Assignments
+
+#### spec-only (8 waves)
+
+| Wave | Task | Role | Notes |
+|------|------|------|-------|
+| 1 | RESEARCH-001 | analyst | Domain research |
+| 2 | DRAFT-001 | writer | Product brief |
+| 3 | DRAFT-002 | writer | Requirements PRD |
+| 4 | CHECKPOINT-001 | supervisor | Brief↔PRD consistency |
+| 5 | DRAFT-003 | writer | Architecture |
+| 6 | DRAFT-004 | writer | Epics & stories |
+| 7 | CHECKPOINT-002 | supervisor | Full spec consistency |
+| 8 | QUALITY-001 | reviewer | Readiness gate |
+
+#### impl-only (5 waves)
+
+| Wave | Task | Role | Notes |
+|------|------|------|-------|
+| 1 | PLAN-001 | planner | Implementation planning |
+| 2 | CHECKPOINT-003 | supervisor | Plan↔input alignment |
+| 3 | IMPL-001(~N) | executor | Implementation (parallel if multiple) |
+| 4 | TEST-001 + REVIEW-001 | tester + reviewer | Validation (parallel) |
+| 5 | (reserved) | — | For IMPROVE tasks if review finds issues |
+
+#### full-lifecycle (13 waves)
+
+spec-only waves 1-8 + user approval checkpoint + impl-only waves shifted.
+
+| Wave | Task | Role |
+|------|------|------|
+| 1-8 | (spec-only pipeline) | (see above) |
+| 9 | PLAN-001 | planner |
+| 10 | CHECKPOINT-003 | supervisor |
+| 11 | IMPL-001(~N) | executor |
+| 12 | TEST-001 + REVIEW-001 | tester + reviewer |
+| 13 | (reserved) | — |
+</csv_schema>
+
+<invariants>
+1. **Start Immediately**: First action is session initialization
+2. **Wave Order is Sacred**: Execute waves sequentially
+3. **CSV is Source of Truth**: Master tasks.csv holds all state
+4. **Column Separation Rule**: Input and output_schema columns MUST NOT share names
+5. **Context Propagation**: prev_context from master CSV findings
+6. **Discovery Board is Append-Only**: Never clear/modify discoveries.ndjson
+7. **Cascading Skip on Failure**: Failed tasks cascade to dependents
+8. **Cleanup Temp Files**: Delete wave-N.csv and wave-N-results.csv after merge
+9. **Checkpoint Pause**: After CHECKPOINT waves, pause for user approval (skip if -y)
+10. **DO NOT STOP**: Continuous execution between checkpoints
+11. **Role Files are Authoritative**: Agents read roles/{role}/role.md for domain logic
+</invariants>
+
+<state_machine>
+
+<states>
+S_PARSE      — Parse arguments, detect pipeline               PERSIST: —
+S_PIPELINE   — Select and validate pipeline                    PERSIST: —
+S_CSV_GEN    — Generate tasks.csv from pipeline definition     PERSIST: tasks.csv
+S_WAVE_{N}   — Execute wave N                                  PERSIST: findings in master CSV
+S_CHECKPOINT — User approval after checkpoint wave             PERSIST: —
+S_AGGREGATE  — Generate report, export results                 PERSIST: context.md, results.csv
+</states>
+
+<transitions>
+S_PARSE → S_PIPELINE      WHEN: new session
+S_PARSE → S_WAVE_{N}      WHEN: --continue (resume at first pending wave)
+S_PIPELINE → S_CSV_GEN
+S_CSV_GEN → S_WAVE_1
+
+S_WAVE_{N} → S_CHECKPOINT     WHEN: wave N was a CHECKPOINT task
+S_WAVE_{N} → S_WAVE_{N+1}     WHEN: wave N not checkpoint, more waves
+S_WAVE_{N} → S_AGGREGATE      WHEN: last wave complete
+
+S_CHECKPOINT → S_WAVE_{N+1}   WHEN: user approves or -y
+S_CHECKPOINT → S_WAVE_{N}     WHEN: user requests revision (re-run checkpoint)
+S_CHECKPOINT → S_AGGREGATE    WHEN: user aborts
+</transitions>
+
+<actions>
+
+### Session Initialization (S_PARSE)
+
+```
+Parse from $ARGUMENTS:
+  AUTO_YES        ← --yes | -y
+  continueMode    ← --continue
+  maxConcurrency  ← --concurrency | -c N  (default: 3)
+  pipelineFlag    ← --pipeline spec-only|impl-only|full-lifecycle
+  noSupervision   ← --no-supervision (skip CHECKPOINT tasks)
+  taskDescription ← remaining text
+
+Derive:
+  dateStr        ← UTC+8 YYYYMMDD
+  slug           ← first 3 meaningful words, kebab-case
+  sessionId      ← "{dateStr}-tlv4-{slug}"
+  sessionFolder  ← ".workflow/.csv-wave/{sessionId}"
+  skillRoot      ← resolve path to this skill directory
+
+mkdir -p {sessionFolder}/{spec,plan,artifacts}
+```
+
+### Pipeline Selection (S_PIPELINE)
+
+1. If `--pipeline` flag → use specified pipeline
+2. Else scan task description keywords against [specs/pipelines.md](specs/pipelines.md) §1:
+   - spec/design/document/requirements → `spec-only`
+   - implement/build/fix/code → `impl-only`
+   - full/lifecycle/end-to-end → `full-lifecycle`
+   - Ambiguous → `request_user_input`
+3. Validate pipeline selection
+
+### CSV Generation (S_CSV_GEN)
+
+1. Load pipeline definition from [specs/pipelines.md](specs/pipelines.md)
+2. **Supervision opt-out**: if `--no-supervision`, filter out all CHECKPOINT-* tasks from pipeline. Adjust deps: tasks that depended on CHECKPOINT tasks now depend on the CHECKPOINT's upstream instead.
+3. For each task in pipeline, build CSV row:
+   - `id`: from task registry (RESEARCH-001, DRAFT-001, etc.)
+   - `description`: PURPOSE/TASK/EXPECTED/CONSTRAINTS with session-specific context
+   - `role`: from registry
+   - `pipeline_phase`: from registry
+   - `deps`: from pipeline dependency chain
+   - `context_from`: same as deps (upstream findings needed)
+   - `wave`: from pipeline wave assignment table (recalculate if checkpoints removed)
+4. Initialize lifecycle columns: `status=pending`, empty `findings`/`files_modified`/`quality_score`/`error`
+5. Write `tasks.csv`
+6. Write empty `discoveries.ndjson`
+7. User validation (skip if `-y`): display pipeline, task count, wave structure
+
+### Wave Execution Engine (S_WAVE_{N})
+
+For each wave N:
+
+#### Step 1: Skip check
+- If all tasks in wave N are already completed/skipped → skip to next wave
+- If any dep is failed/blocked → cascade skip
+
+#### Step 2: Build prev_context
+For each task in wave N:
+- Read `context_from` task IDs → extract their `findings` from master CSV
+- Concatenate as `prev_context`
+
+#### Step 3: Write wave-{N}.csv
+Extract wave N rows + add `prev_context` column.
+
+#### Step 4: Execute
+
+```javascript
+spawn_agents_on_csv({
+  csv_path: `${sessionFolder}/wave-${N}.csv`,
+  id_column: "id",
+  instruction: buildLifecycleInstruction(sessionFolder, skillRoot),
+  max_concurrency: maxConcurrency,
+  max_runtime_seconds: 3600,
+  output_csv_path: `${sessionFolder}/wave-${N}-results.csv`,
+  output_schema: {
+    id,
+    result_status,    // completed | failed | blocked
+    findings,         // key findings (max 500 chars)
+    files_modified,   // semicolon-separated paths
+    quality_score,    // 0-100
+    error             // error message
+  }
+})
+```
+
+#### Step 5: Merge + Cleanup
+1. Map `result_status` → master `status`
+2. Copy `findings`, `files_modified`, `quality_score`, `error`
+3. Delete `wave-{N}.csv` AND `wave-{N}-results.csv`
+4. Cascade skip failed task dependents
+
+#### Step 6: Checkpoint handling
+If wave N was a CHECKPOINT task:
+- Read supervisor's `findings` and `quality_score`
+- If quality_score >= 80 (pass) → continue
+- If 60-79 (review) → display warnings, continue (or pause if not -y)
+- If < 60 (fail) → pause for user: Revise / Override / Abort
+
+### Instruction Builder
+
+```
+You are a team-lifecycle-v4 agent executing a software development task.
+
+## Your Identity
+- Role: read from your CSV row 'role' column
+- Task: read from your CSV row 'description' column (PURPOSE/TASK/EXPECTED/CONSTRAINTS)
+- Phase: read from your CSV row 'pipeline_phase' column
+
+## Role Definition
+Read your detailed role instructions from:
+  {skillRoot}/roles/{role}/role.md
+
+Follow the execution protocol defined in your role.md (Phase 2-4 structure).
+
+For roles with commands/ subdirectory, load the relevant command file based on your task type.
+
+## Context
+- Session: {sessionFolder}
+- Discovery board: {sessionFolder}/discoveries.ndjson (read before work, append findings)
+- Previous context: read your CSV row 'prev_context' column
+- Templates: {skillRoot}/templates/ (for writer role)
+- Quality gates: {skillRoot}/specs/quality-gates.md
+
+## Quality Contract
+1. Verify all outputs exist before reporting completed
+2. Include quality_score (0-100) based on specs/quality-gates.md criteria
+3. Retry on verification failure (max 2 retries)
+4. Report blocked if still failing after retries
+
+## Discovery Protocol
+Write task output to {sessionFolder}/discoveries/{task_id}.json:
+{
+  "task_id": "<id>", "role": "<role>", "status": "<result_status>",
+  "findings": "<summary>", "quality_score": <N>,
+  "artifacts_produced": ["<paths>"], "files_modified": ["<paths>"]
+}
+
+Append to {sessionFolder}/discoveries.ndjson:
+{"ts":"<ISO>","worker":"<id>","type":"<pipeline_phase>","data":{...}}
+
+## Output
+Return via output_schema:
+- result_status: completed | failed | blocked
+- findings: key findings (max 500 chars)
+- files_modified: semicolon-separated paths
+- quality_score: 0-100 per quality-gates.md
+- error: error message if not completed
+```
+
+### Results Aggregation (S_AGGREGATE)
+
+1. Export `results.csv`
+2. Generate `context.md`:
+   - Pipeline type, task count, completed/failed/skipped
+   - Per-wave results with quality scores
+   - Checkpoint verdicts
+   - Files modified
+   - Discovery summary
+   - Next steps (if spec-only: suggest impl-only; if impl: suggest deploy)
+3. Display completion report
+
+</actions>
+</state_machine>
+
+<discovery_board>
+
+| Type | Dedup Key | Data |
+|------|-----------|------|
+| research_finding | topic | {topic, evidence, implications} |
+| spec_artifact | doc_type | {doc_type, path, summary, quality_score} |
+| architecture_decision | subject | {subject, choice, rationale, alternatives[]} |
+| implementation_note | module | {module, pattern, files[], notes} |
+| test_result | suite | {suite, pass_rate, failures[], coverage} |
+| review_finding | dimension | {dimension, severity, location, recommendation} |
+| checkpoint_verdict | checkpoint_id | {id, score, checks[], verdict} |
+
+Protocol: read before work, append-only, dedup by type+key.
+</discovery_board>
+
+<error_codes>
+
+| Condition | Recovery |
+|-----------|----------|
+| Pipeline detection ambiguous | request_user_input for pipeline selection |
+| RESEARCH-001 failed | Abort spec pipeline (no context for drafting) |
+| DRAFT task failed | Skip subsequent DRAFTs, proceed to checkpoint |
+| CHECKPOINT verdict: block | Pause for user (Revise/Override/Abort) |
+| PLAN-001 failed | Abort impl pipeline |
+| IMPL task failed | Cascade skip to TEST + REVIEW |
+| Session not found (--continue) | Error with available sessions |
+| Role file missing | Error with role registry |
+| Quality score < 60 after revision | Escalate to user |
+</error_codes>
+
+<success_criteria>
+- [ ] Pipeline selected and validated
+- [ ] tasks.csv generated with correct wave assignments
+- [ ] Each wave executed via spawn_agents_on_csv
+- [ ] Agents read role.md for domain-specific logic
+- [ ] Column separation rule maintained
+- [ ] wave-N.csv and wave-N-results.csv deleted after merge
+- [ ] Checkpoint waves pause for user approval (unless -y)
+- [ ] Quality scores reported per task
+- [ ] discoveries.ndjson append-only
+- [ ] results.csv and context.md generated
+- [ ] Session resumable via --continue
+- [ ] Spec artifacts written to spec/ directory
+- [ ] Plan artifacts written to plan/ directory
+</success_criteria>