agentleFS
DeL-TaiseiOzakiDeL-TaiseiOzaki/claude-code-orchestra/.claude/skills/plan/SKILL.md

plan

Skill199 starsChanged 2 months agoMITView on GitHub

Create a detailed implementation plan for a feature or task. Use when user wants to plan before coding.

184 other projects use a version of this, 2 of them edited.

Adopted by

Most recently taken up by

Used with

Projects with it also use heartmula, webhook-subscriptions and 6 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • math-incSkill

    heartmula

    Set up and run HeartMuLa, the open-source music generation model family (Suno-like). Generates full songs from lyrics + tags with multilingual support.

    Also in 9 projects that use this

  • Tommy-ywSkill

    webhook-subscriptions

    Create and manage webhook subscriptions for event-driven agent activation, or for direct push notifications (zero LLM cost). Use when the user wants external services to trigger agent runs OR push notifications to chats.

    Also in 5 projects that use this

  • math-incSkill

    minecraft-modpack-server

    Set up a modded Minecraft server from a CurseForge/Modrinth server pack zip. Covers NeoForge/Forge install, Java version, JVM tuning, firewall, LAN config, backups, and launch scripts.

    Also in 5 projects that use this

  • john-data-chenSkill

    touchdesigner-mcp

    Control a running TouchDesigner instance via twozero MCP — create operators, set parameters, wire connections, execute Python, build real-time visuals. 36 native tools.

    Also in 6 projects that use this

  • ibahgatSkill

    context-optimize

    Improve context efficiency and cache stability for long or complex sessions.

    Also in 3 projects that use this

  • +3 more from math-inc

The file

We split this skill into the rules it gives an agent

8 rules. 0 are common, 4 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 168 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Create an implementation plan for $ARGUMENTS.

    line 9
  2. Very specific:

    If .claude/docs/plans/{slug}.md already exists, the plan is a revision of that document — read it first and say so, rather than silently …

    line 37
  3. Very specific:

    Ask, do not assume. Unresolved items belong in ### Open Questions, not in a guessed step.

    line 48
  4. Distinctive:

    Write the prompt body to a file, then invoke the shared wrapper (the directory is not created by the wrapper before it reads the prompt, so …

    line 85
  5. Very specific:

    Read the answer from the JSON output's response_file.

    line 117
  6. Very specific:

    Write the plan to the plan_doc path from step 0 (.claude/docs/plans/{slug}.md) using this shape.

    line 125
  7. Distinctive:

    Ask questions at planning stage for unclear points

    line 233
  8. Distinctive:

    Don't over-detail (adjust during implementation)

    line 234

“Common” is measured: kept by at least 6 in 10 of the 168 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Create Implementation Plan

Create an implementation plan for $ARGUMENTS.

A plan is a file, not a transcript. The plan produced here is written to a resolved path, validated against a document contract, reviewed by Codex, and approved by the user — in that order, and all four. A plan that lives only in the conversation cannot be handed to /team-execute, cannot survive a /checkpointing compaction, and cannot be re-read next session.

Planning Process

0. Resolve the Workspace (always first)

Resolve this plan's slug and output path once. The title becomes the file name, so give it a short English descriptor — not the user's raw wording, which the Language Protocol keeps out of paths:

python3 .claude/skills/_shared/workspace.py \
  --skill plan --title "<short English title>" --create

The JSON on stdout carries slug and paths.plan_doc (.claude/docs/plans/{slug}.md). Every {slug} and every output path below — and the /team-execute handoff in step 6 — MUST come from this JSON verbatim, never be re-derived by hand: two hand-derived slugs are exactly how a plan and its executor's work logs drift apart. Exit codes: 0 resolved (and created); 1 bad arguments.

If .claude/docs/plans/{slug}.md already exists, the plan is a revision of that document — read it first and say so, rather than silently overwriting it.

1. Requirements Analysis

First clarify with the user:

  • Purpose: What to achieve
  • Scope: What to include, what to exclude
  • Constraints: Technical, time, dependencies

Ask, do not assume. Unresolved items belong in ### Open Questions, not in a guessed step.

2. Current State Investigation

Reading the codebase for a plan is broad-context work: delegate it rather than consuming orchestrator context.

Task tool:
  subagent_type: "general-purpose-opus"
  prompt: |
    Investigate this codebase for a planned change: {purpose from step 1}

    Report, with file:line evidence for every claim:
    1. Related existing code — which modules already do something similar
    2. Files that will need to change, and why each one
    3. Libraries and patterns the change should reuse (project conventions)
    4. Existing tests that cover the affected area, and gaps in that coverage
    5. Constraints or invariants the change must not break

    Use Glob, Grep, and Read. Do not implement anything.
    Return the findings; do not summarize away the file paths.

The returned findings are the source of ### Scope and ### Risks & Considerations in the plan document. A plan whose Scope names no concrete file paths is a plan written without reading the code — send the investigation back rather than proceeding.

3. Break Down Implementation Steps

.claude/rules/codex-delegation.md names "you need a step-by-step implementation plan" and "design/architecture decisions are involved" as explicit Codex triggers, so step decomposition is a MANDATORY Codex consult — the same rule /feature follows for the same artifact class.

Write the prompt body to a file, then invoke the shared wrapper (the directory is not created by the wrapper before it reads the prompt, so create it first):

mkdir -p .claude/logs/codex
# write the prompt body to .claude/logs/codex/prompt-plan-{slug}-steps.md, then:
python3 .claude/skills/_shared/codex_consult.py \
  --prompt-file .claude/logs/codex/prompt-plan-{slug}-steps.md \
  --label plan-{slug}-steps --sandbox read-only

Prompt body:

Objective: Decompose this change into an ordered, independently testable implementation plan.
Context:
- Purpose: {purpose from step 1}
- Scope in / out: {from step 1}
- Constraints: {from step 1}
- Current state: {findings from step 2, with file paths}
Constraints:
- Order steps by dependency (what must exist before what)
- Each step is independently testable; name its verification
- Put high-risk and high-uncertainty steps first
- Keep steps small; do not over-detail work that will be re-decided during implementation
Output format:
## Implementation Steps (ordered by dependency)
## Verification per Step
## Risks and Mitigations
## Open Questions

Read the answer from the JSON output's response_file. Exit codes: 0 the call succeeded — read response_file; 1 bad args; 2 codex CLI not on PATH; 3 codex exited non-zero or timed out — inspect error and stderr_file before retrying or escalating. Codex proposes the decomposition; deciding which steps are the right steps stays yours.

4. Output Format

Write the plan to the plan_doc path from step 0 (.claude/docs/plans/{slug}.md) using this shape. The five ### sections are the plan-doc document contract — renaming or dropping one is a validation failure in step 5:

## Implementation Plan: {Title}

### Purpose
{1-2 sentences}

### Scope
- New files: {list}
- Modified files: {list}
- Dependencies: {list}

### Implementation Steps

#### Step 1: {Title}
- [ ] {Specific task}
- [ ] {Specific task}
**Verification**: {Completion criteria for this step}

#### Step 2: {Title}
...

### Risks & Considerations
- {Potential issues and mitigations}

### Open Questions
- {Items to clarify before implementation}

5. Completion Gates

A plan is done when all four gates below have passed. They check different things and none of them substitutes for another: the shape gate proves the document has its sections, never that the plan is any good.

Gate 1 — shape. The document contract:

python3 .claude/skills/_shared/validate_doc.py \
  --contract plan-doc --file .claude/docs/plans/{slug}.md

Exit 0 the required sections are present; 1 bad args or the file is unreadable; 2 a required section is missing — read sections_missing in the JSON and add the section. Exit 0 means the plan has the right shape. It says nothing about whether the steps are correct, complete, or ordered sensibly.

Gate 2 — the artifact exists and is not empty:

python3 .claude/skills/_shared/workspace.py --skill plan --slug {slug} --verify

Exit 0 the plan document exists and is non-empty; exit 2 it is missing or effectively empty — the plan was never written, so do not report a plan.

Gate 3 — adequacy (MANDATORY Codex validation). Same rigour /feature applies to its implementation plan. Write the body below to .claude/logs/codex/prompt-plan-{slug}-validate.md and consult with --sandbox read-only:

Objective: Validate this implementation plan for completeness, correctness, and risk.
Context:
- Purpose and scope: {from step 1}
- Current state: {findings from step 2}
- Implementation plan: contents of .claude/docs/plans/{slug}.md
Constraints:
- Check for missing edge cases or error handling
- Verify the step order is actually dependency-correct
- Ensure each step's stated verification would really detect a failure
- Identify integration risks and convention violations
- Check that nothing in Open Questions is in fact a blocker disguised as a question
Output format:
## Validation Result (PASS / NEEDS_REVISION)
## Missing Coverage
## Ordering Problems
## Integration Risks
## Revised Steps (if NEEDS_REVISION)

On NEEDS_REVISION, revise the plan document and re-run gates 1 and 3 before continuing. How to revise is judgment; re-validating is not optional.

Gate 4 — user approval. Present the plan (path included) and ask before any implementation starts. No script can decide that a plan is good enough to build.

6. Handoff

Report the resolved path so the next skill can consume it:

Plan written to .claude/docs/plans/{slug}.md (slug: {slug}) — Codex: PASS
Next: /team-execute with slug {slug}

/team-execute resolves its own workspace from that same slug and reads the plan from .claude/docs/plans/{slug}.md. Hand over the path and the slug; do not paste the plan back into the conversation as the source of truth, and do not invoke another skill from here — this is a handoff the user makes.

Notes

  • Plans should be at actionable granularity
  • Include verification method for each step
  • Ask questions at planning stage for unclear points
  • Don't over-detail (adjust during implementation)
  • Deciding the steps, the risks, and what belongs in Open Questions is judgment and stays prose; only the artifact's path, shape, and gates are mechanical

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.