메뉴
Drive multi-step plan implementation with focused context packets. USE WHEN step prepare OR step check OR step execute OR step validate OR step fix OR step done OR implementation workflow OR plan packet.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | ManageImpStep |
| description | Drive multi-step plan implementation with focused context packets. USE WHEN step prepare OR step check OR step execute OR step validate OR step fix OR step done OR implementation workflow OR plan packet. |
| allowed-tools | Read |
Structured workflow for implementing multi-step plans. Creates focused "step packets" with all context needed for implementation, tracks status directly in plan files.
{phase-id}/{step-id} via HTML comments in plan<!-- id: step-id status: in-progress|done --> in plan file| Workflow | Trigger | File |
|---|---|---|
| Prepare | /ManageImpStep prepare <plan> <design-doc> [step-id] [--auto-check] [--debug] | Workflows/Prepare.md |
| Check | /ManageImpStep check <packet> [--orchestrate] [--dc-pass] [--loop] [--double-check] [--max-cycles N] [--debug] | Workflows/Check.md |
| Execute | /ManageImpStep execute <packet> | Workflows/Execute.md |
| Validate | /ManageImpStep validate <packet> | Workflows/Validate.md |
| Fix | /ManageImpStep fix <packet> [extra-feedback] | Workflows/Fix.md |
| Done | /ManageImpStep done <packet> [--no-commit] | Workflows/Done.md |
PREPARE --auto-check → CHECK --orchestrate (background Task agent)
│
├── single-pass iterations (background Tasks)
└── dc-pass evaluation (background Task)
↓
PREPARE → CHECK (optional) → EXECUTE → VALIDATE ──→ DONE
│ ↑ ↓
└── creates packet └── FIX
| State | HTML Comment |
|---|---|
| Todo | <!-- id: step-id --> (no status) |
| Prepared | <!-- id: step-id status: prepared --> (packet ready) |
| In Progress | <!-- id: step-id status: in-progress --> |
| Done | <!-- id: step-id status: done --> |
Given plan plans/foo/bar.md and step phase-id/step-id:
→ Packet: plans/foo/bar-steps/phase-id/step-id.md
Example 1: Start implementing next step
User: "/ManageImpStep prepare plans/personal-finance/personal-finance.md docs/personal-finance/README.md"
→ Invokes Prepare workflow
→ Finds first todo step, creates packet, marks in-progress
→ Output: "Prepared step cli-foundation/config-loading. Packet: plans/.../config-loading.md"
Example 2: Prepare specific step
User: "/ManageImpStep prepare plans/myplan.md docs/myplan/README.md import-session/excel-parser"
→ Invokes Prepare workflow for specific step
→ Creates packet for excel-parser step
Example 3: Full implementation cycle
User: "/ManageImpStep prepare ..." → creates packet
User: "/ManageImpStep check <packet>" → verifies packet completeness (single-pass)
User: "/ManageImpStep execute <packet>" → implements the step
User: "/ManageImpStep validate <packet>" → runs tests, checks criteria
User: "/ManageImpStep done <packet>" → marks done, commits (default)
Example 4: Fix failed validation
User: "/ManageImpStep validate <packet>" → FAIL, writes findings file
User: "/ManageImpStep fix <packet>" → reads findings, applies targeted TDD fixes
User: "/ManageImpStep validate <packet>" → PASS, deletes findings file
User: "/ManageImpStep done <packet>" → marks done, commits
Example 5: Single step with auto-check (background)
User: "/ManageImpStep prepare plans/myplan.md docs/myplan/README.md --auto-check"
→ Normal single-step prepare
→ After packet creation, spawns background Task agent running Check --orchestrate
→ Reports immediately with background output file path
→ Check runs asynchronously: single-pass iterations + dc-pass evaluation
Example 6: Check with orchestrator (iterative + double-check)
User: "/ManageImpStep check plans/myplan-steps/phase/step.md --orchestrate"
→ Phase A: spawns single-pass iterations as background Tasks until convergence or 10 iterations
→ Phase B: spawns dc-pass evaluation as background Task (7 dimensions)
→ If any dimension WEAK/MISSING: fix, restart Phase A (max 3 cycles)
→ Updates frontmatter with check-confidence: double-checked
Example 7: Check single-pass (manual, one iteration)
User: "/ManageImpStep check plans/myplan-steps/phase/step.md"
→ One pass: read source docs → analyze gaps → fix → report
→ No looping, no double-check — fast manual review
Example 7B: Legacy flags (backwards compat)
User: "/ManageImpStep check plans/myplan-steps/phase/step.md --loop --double-check"
→ Maps to --orchestrate internally
→ Same behavior as Example 6
Example 8: Resuming a prepared step
User: "/ManageImpStep prepare plans/myplan.md docs/myplan/README.md"
→ First non-done step has status: prepared
→ Sets it to in-progress, skips packet generation
→ Output: "Using pre-generated packet for step-id. Packet: plans/.../step-id.md"
Example 9: Parallel preparation (multiple Claude instances)
# Run in separate Claude Code instances:
Instance 1: "/ManageImpStep prepare plans/myplan.md docs/design.md phase/step-a --auto-check"
Instance 2: "/ManageImpStep prepare plans/myplan.md docs/design.md phase/step-b --auto-check"
→ Each instance has full Skill tool → hooks fire correctly
→ Use explicit step-id to avoid conflicts
Example 10: Debug mode (visibility into check pipeline)
# Terminal 1: watch the debug log
tail -f ~/.claude/hooks/validators/ManageImpStep/check-debug.log
# Terminal 2: run prepare with debug
User: "/ManageImpStep prepare plans/myplan.md docs/design.md --auto-check --debug"
→ Debug log shows: [prepare] Auto-check spawned, [orchestrate] Started, [single-pass] iterations, [dc-pass] scores
→ Cross-reference timestamps with packet-structure-validator.log to verify hooks fired
PRD lifecycle from vague idea to structured spec. USE WHEN design product OR write PRD OR review PRD OR structure PRD OR restructure PRD OR validate PRD OR check consistency OR verify PRD OR run checks OR one-pager OR "too complex" OR complexity signals OR split PRD OR process findings OR fix findings OR go through findings.
Create, validate, and optimize Claude Code skills. USE WHEN user says "skill" + intent (create, make, build, want, need, help me with) OR "workflow" + intent (create, make, add, new) OR "workflow for/to" OR "automate [task]" OR mentions skill structure, validate, audit OR optimize workflow OR improve workflow OR analyze session OR workflow delta.
Extract insights and patterns from YouTube coding videos and their GitHub repos. USE WHEN analyze video OR learn from video OR extract lessons OR video insights OR IndyDevDan OR agentic engineer video.
Git practices for this project. USE WHEN committing code OR creating commits OR git operations. Enforces clean git history.
Create, modify, review, and harden agentic prompts/slash commands. USE WHEN create prompt OR new command OR slash command OR modify prompt OR update workflow OR improve prompt OR edit prompt OR prompt structure OR workflow prompt OR skill workflow OR add workflow to skill OR workflow file OR write workflow OR review prompt OR harden prompt OR add validation OR add hooks OR self-validation OR validate agent OR add self-validation. Guides through 7 prompt levels with proper sections.
Expert guidance for creating, writing, and refining Claude Code Skills. Use when working with SKILL.md files, authoring new skills, improving existing skills, or understanding skill structure and best practices.