kochetkov-ma/brewcode:plan

Create Plan — [task-dir or SPEC path or plan file]

<instructions>

/brewcode:plan Instructions

ROLE: Plan Creator | OUTPUT: PLAN.md + phases/*.md + KNOWLEDGE.jsonl + artifacts/ + backup/

Input Detection

| Input | Action | |-------|--------| | Path to {TS}_{NAME}_task/ dir | Read SPEC.md from it | | Path to SPEC.md file | Derive task dir from parent | | .claude/plans/LATEST.md or plan file | Plan Mode: parse plan, create task dir, skip SPEC | | Empty | Check .claude/TASK.md quick ref for latest task dir |

Flag Parsing

Parse $ARGUMENTS for flags BEFORE input detection:

| Flag | Effect | |------|--------| | -n, --noask | Skip all user questions, auto-approve defaults |

Strip flag from $ARGUMENTS. Remaining text = path.

Workflow (SPEC input)

0. Check Adapted Templates (REQUIRED FIRST)

   EXECUTE using Bash tool:

   TMPL_DIR=".claude/tasks/templates"
   PLUGIN_TMPL="$BC_PLUGIN_ROOT/skills/setup/templates"
   echo "--- Project templates ---"
   test -f "$TMPL_DIR/PLAN.md.template" && echo "PLAN.md.template OK" || echo "PLAN.md.template MISSING"
   test -f "$TMPL_DIR/phase.md.template" && echo "phase.md.template OK" || echo "phase.md.template MISSING"
   test -f "$TMPL_DIR/phase-verify.md.template" && echo "phase-verify.md.template OK" || echo "phase-verify.md.template MISSING"
   test -f "$TMPL_DIR/phase-final-review.md.template" && echo "phase-final-review.md.template OK" || echo "phase-final-review.md.template MISSING"
   test -f "$TMPL_DIR/phase-fix.md.template" && echo "phase-fix.md.template OK" || echo "phase-fix.md.template MISSING"
   echo "--- Plugin fallback templates ---"
   test -f "$PLUGIN_TMPL/PLAN.md.template" && echo "PLAN.md.template FALLBACK OK" || echo "PLAN.md.template FALLBACK MISSING"
   test -f "$PLUGIN_TMPL/phase.md.template" && echo "phase.md.template FALLBACK OK" || echo "phase.md.template FALLBACK MISSING"
   test -f "$PLUGIN_TMPL/phase-verify.md.template" && echo "phase-verify.md.template FALLBACK OK" || echo "phase-verify.md.template FALLBACK MISSING"
   test -f "$PLUGIN_TMPL/phase-final-review.md.template" && echo "phase-final-review.md.template FALLBACK OK" || echo "phase-final-review.md.template FALLBACK MISSING"
   test -f "$PLUGIN_TMPL/phase-fix.md.template" && echo "phase-fix.md.template FALLBACK OK" || echo "phase-fix.md.template FALLBACK MISSING"
   

   Template resolution order:

   1. Project templates: .claude/tasks/templates/{name}.template (adapted by /brewcode:setup)
   2. Plugin fallback: $BC_PLUGIN_ROOT/skills/setup/templates/{name}.template

   STOP if BOTH locations MISSING for any template -- Run /brewcode:setup first to get v3 templates. If project templates missing but plugin fallback exists: WARN "Re-run /brewcode:setup for v3 project-adapted templates. Using plugin defaults."

1. Read SPEC

   • Resolve input path per Input Detection table
   • Read SPEC.md from task directory
   • Extract: goal, requirements, analysis, context files, risks, decisions

2. Scan Project for Reference Examples

   • Find 1-2 canonical files per expected phase type (controller, service, test, etc.)
   • These become Reference Examples (R1, R2...) in PLAN.md

Dynamic Agent Resolution

Before assigning agents to phases, check for project team agents:

1. If .claude/teams/ exists — read team.md for agent roster with domains
2. If .claude/agents/ has project agents — list available
3. Match agent domain to phase task area
4. Priority: team agent > project agent > plugin agent > system agent
5. If agent refuses (Task Acceptance Protocol) — re-delegate to suggested colleague (max 2 retries)

Always fall back to plugin agents when no project agents match the task domain.

3. Generate Phase Breakdown (5-12 phases)

   Based on SPEC analysis and project structure:

   • Each phase = one logical unit of work
   • Dependencies between phases identified
   • Agents assigned per phase
   • Verification phases (NV) after each execution phase
   • Final Review (FR) as last phase

4. Present Phases to User (AskUserQuestion)

   If --noask: Skip. Auto-approve all phases.

   Otherwise: Use AskUserQuestion to present the proposed phase split:

   • Phase count and descriptions
   • Agent assignments
   • Dependency chain
   • User can approve, adjust, or request changes

5. Generate Artifacts

   Read templates per resolution order from Step 0 (project first, plugin fallback second).

   5.1 Create directory structure

   .claude/tasks/{TS}_{NAME}_task/
   ├── phases/           <-- NEW: individual phase files
   ├── artifacts/
   ├── backup/
   ├── KNOWLEDGE.jsonl   (0-byte empty file)
   └── PLAN.md           (slim, with Phase Registry)
   

   EXECUTE using Bash tool:

   TASK_DIR=".claude/tasks/{TS}_{NAME}_task"
   mkdir -p "$TASK_DIR/phases" "$TASK_DIR/artifacts" "$TASK_DIR/backup"
   touch "$TASK_DIR/KNOWLEDGE.jsonl"
   

   5.2 Generate phase files (from templates)

   For EACH execution phase (1, 2, 3, ...):

   • Read phase.md.template
   • Fill placeholders with SPEC-derived content:
     • {PHASE_NUM} -- phase number
     • {PHASE_NAME} -- descriptive name (kebab-case for filename, Title Case for heading)
     • {AGENT} -- assigned agent
     • {AGENT_ROLE} -- one-line agent role description
     • {OBJECTIVE} -- concrete objective from SPEC analysis
     • {CONTEXT_FILES} -- table rows: files the agent needs to read/modify
     • {REFERENCES} -- table rows: reference examples, docs, existing patterns
     • {TASK_LIST} -- numbered task list with specific, actionable items
     • {CONSTRAINTS} -- project-specific constraints (from SPEC + project rules)
     • {EXIT_CRITERIA} -- measurable exit criteria (builds, tests pass, lint clean, etc.)
     • {ARTIFACT_DIR} -- e.g., 1-1e (Phase 1 Execution, iter 1)
     • {ADDITIONAL_ARTIFACTS} -- files created/modified by this phase
   • Write to: phases/{N}-{name}.md (e.g., phases/1-create-entity.md)

   For EACH verification phase:

   • Read phase-verify.md.template
   • Fill placeholders:
     • {PHASE_NUM} -- matches the execution phase being verified
     • {PHASE_NAME} -- same name as the execution phase
     • {VERIFY_AGENT} -- tester or reviewer
     • {FILES_TO_REVIEW} -- files created/modified by the execution phase
     • {VERIFICATION_CHECKLIST} -- checklist items derived from exit criteria of execution phase
     • {AGAINST_REFERENCES} -- reference examples to compare against
     • {ARTIFACT_DIR} -- e.g., 1-1v (Phase 1 Verification, iter 1)
     • {AGENT} -- same as {VERIFY_AGENT}
   • Write to: phases/{N}V-verify-{name}.md

   For Final Review:

   • Read phase-final-review.md.template
   • Fill placeholders:
     • {FR_AGENTS} -- list of review agents (typically reviewer + tester + architect)
     • {COMPLETION_CRITERIA} -- from SPEC.md goals/decisions, copied to PLAN.md Completion Criteria
     • {REVIEW_CHECKLIST} -- comprehensive checklist covering all phases
     • {FILES_CHANGED} -- aggregate of all files created/modified across all phases
     • {ARTIFACT_DIR} -- FR-1e (Final Review, iter 1)
   • Write to: phases/FR-final-review.md

   5.3 Generate PLAN.md (slim v3 format)

   Using PLAN.md.template (project-adapted or plugin fallback):

   • Fill Phase Registry table with ALL generated phase files
   • Each row references the corresponding phases/{file}.md
   • Completion Criteria from SPEC.md decisions/goals
   • Agents table from project analysis
   • If .claude/teams/ exists: populate ### Project Agents table from team.md roster
   • Technology Choices from SPEC analysis
   • Role Constraints from project rules

   5.4 Technology Choices

   For each non-trivial choice (library, pattern, approach):

   • Document in PLAN.md under Technology Choices section
   • Include rationale + alternatives considered and rejected
   • Examples: ORM choice, auth library, caching strategy, test framework

6. Quorum Plan Review (3 agents, mixed expertise)

   ONE message with 3 Task calls in PARALLEL:
   
   Task(subagent_type="Plan", prompt="Review PLAN.md and ALL files in phases/ against SPEC requirements. Check Phase Registry completeness and phase file content quality.")
   Task(subagent_type="brewcode:architect", prompt="Review PLAN.md and phases/ files: architecture decisions, technology choices, dependencies between phases, context files accuracy")
   Task(subagent_type="brewcode:reviewer", prompt="Review PLAN.md and phases/ files: exit criteria measurability, task granularity, risks, verification checklist completeness")
   

   Agent prompt template:

   > **Context:** BC_PLUGIN_ROOT is available in your context (injected by pre-task.mjs hook).
   
   Review PLAN at {PLAN_PATH} and phase files at {TASK_DIR}/phases/ against SPEC at {SPEC_PATH}
   Check: Phase Registry matches actual phase files, each phase file has filled content (no unfilled placeholders),
   agent assignments match expertise, dependencies correct, exit criteria measurable, risks mitigated
   Output: list of remarks with rationale
   

   Quorum rule (2/3): Only remarks confirmed by 2+ agents are accepted.

7. Verification Agent (Traceability Check)

   Task(subagent_type="brewcode:reviewer", prompt="
   > **Context:** BC_PLUGIN_ROOT is available in your context (injected by pre-task.mjs hook).
   
   Verify PLAN and phase files cover ALL SPEC requirements:
   - Each item from SPEC > Scope > In has at least one phase FILE in phases/
   - Each requirement from Original Requirements is addressed in a specific phase file's Tasks section
   - Phase Registry in PLAN.md matches actual files in phases/ directory
   Output: traceability matrix (requirement -> phase file) + gaps found")
   

   If gaps found: Add missing phase files AND update Phase Registry in PLAN.md before presenting to user in Step 8.

8. Present Review Results (AskUserQuestion)

   If --noask: Auto-accept all quorum-confirmed remarks. Fix all in PLAN.md and phase files.

   Otherwise: Present confirmed remarks + verification results to user. User approves/rejects each. Fix approved remarks in PLAN.md and phase files.

Workflow (Plan Mode input)

0. Check Templates -- same as SPEC workflow Step 0

1. Read Plan File

   • Parse .claude/plans/LATEST.md or provided plan file
   • Extract structure, goals, steps

2. Create Task Dir + Scan Project

   • Generate timestamp and name slug
   • Create .claude/tasks/{TS}_{NAME}_task/
   • Scan project for context files and reference examples

3. Split into Granular Phases (finer than plan)

   • Each plan item may become 1-3 phases
   • Add verification phases
   • Add Final Review phase

4. Present Phases to User (AskUserQuestion)

   • Same as SPEC workflow step 4 (respects --noask)

5. Generate Artifacts

   • Same as SPEC workflow Step 5 (phases/, PLAN.md, KNOWLEDGE.jsonl, artifacts/, backup/)
   • No SPEC.md in this flow (plan replaces spec)

6. Lightweight Plan Review (2 agents)

   ONE message with 2 Task calls in PARALLEL:
   
   Task(subagent_type="brewcode:architect", prompt="Review PLAN.md at {PLAN_PATH} and phases/ files: architecture decisions, phase dependencies, agent assignments")
   Task(subagent_type="brewcode:reviewer", prompt="Review PLAN.md at {PLAN_PATH} and phases/ files: phase quality, verification criteria, completeness vs source plan")
   

   Rule: Both agents must confirm a remark for it to be accepted (2/2 consensus). Fix confirmed remarks in PLAN.md and phase files before proceeding.

Update Quick Ref (REQUIRED)

Add task link to TOP of .claude/TASK.md (preserve history):

IF .claude/TASK.md exists:
  1. Read existing content
  2. Prepend: ".claude/tasks/{TS}_{NAME}_task/PLAN.md\n---\n"
  3. Append: existing content
ELSE:
  Create with: ".claude/tasks/{TS}_{NAME}_task/PLAN.md"

Validation (REQUIRED)

EXECUTE using Bash tool:

TS_NAME="{TS}_{NAME}"
TASK_DIR=".claude/tasks/${TS_NAME}_task"
echo "=== Structure ==="
test -d "$TASK_DIR" && echo "TASK_DIR" || echo "TASK_DIR MISSING"
test -f "$TASK_DIR/PLAN.md" && echo "PLAN" || echo "PLAN MISSING"
test -f "$TASK_DIR/KNOWLEDGE.jsonl" && echo "KNOWLEDGE" || echo "KNOWLEDGE MISSING"
test -d "$TASK_DIR/artifacts" && echo "ARTIFACTS" || echo "ARTIFACTS MISSING"
test -d "$TASK_DIR/backup" && echo "BACKUP" || echo "BACKUP MISSING"
test -d "$TASK_DIR/phases" && echo "PHASES_DIR" || echo "PHASES_DIR MISSING"
head -1 .claude/TASK.md 2>/dev/null | grep -q "${TS_NAME}" && echo "QUICK_REF" || echo "QUICK_REF MISSING"
echo "=== Phase Registry vs Files ==="
grep -oP 'phases/[^\s|]+\.md' "$TASK_DIR/PLAN.md" | sort -u | while read -r pf; do
  test -f "$TASK_DIR/$pf" && echo "OK $pf" || echo "MISSING $pf"
done
echo "=== Phase files on disk ==="
for f in "$TASK_DIR"/phases/*.md; do
  test -f "$f" && echo "EXISTS $(basename $f)" || echo "NO PHASE FILES"
done

STOP if any MISSING -- Create missing artifacts before proceeding.

Output

# Plan Created

## Detection

| Field | Value |
|-------|-------|
| Arguments | `{received args}` |
| Input Type | `{SPEC path or Plan Mode file}` |
| Noask | `{yes or no}` |

## Files Created
- PLAN: .claude/tasks/{TS}_{NAME}_task/PLAN.md
- PHASES: .claude/tasks/{TS}_{NAME}_task/phases/ ({count} files)
- KNOWLEDGE: .claude/tasks/{TS}_{NAME}_task/KNOWLEDGE.jsonl
- ARTIFACTS: .claude/tasks/{TS}_{NAME}_task/artifacts/
- BACKUP: .claude/tasks/{TS}_{NAME}_task/backup/
- QUICK REF: .claude/TASK.md (task added to top, history preserved)

## Phase Files
| File | Type | Agent |
|------|------|-------|
| phases/1-{name}.md | Execution | {agent} |
| phases/1V-verify-{name}.md | Verification | {agent} |
| ... | ... | ... |
| phases/FR-final-review.md | Final Review | reviewer+tester+architect |

## Next Step

> Copy the command below first, then clear context and paste it.

1. Clear context: type `/clear` and press Enter
2. Run (paste copied command):
\`\`\`
/brewcode:start .claude/tasks/{TS}_{NAME}_task/PLAN.md
\`\`\`

</instructions>