plan
Builds a phase plan (goal, progress, research, Opus plan review). Use for "plan", "plan this", "adapt this plan", "planejar", "crie o plano", "monte a fase", or a pasted outside plan. Small idea becomes a 1-part plan.
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- plan
- description
- Builds a phase plan (goal, progress, research, Opus plan review). Use for "plan", "plan this", "adapt this plan", "planejar", "crie o plano", "monte a fase", or a pasted outside plan. Small idea becomes a 1-part plan.
- allowed-tools
- Bash(sh "${CLAUDE_PLUGIN_ROOT}/scripts/py" scripts/guarda.py info)
- argument-hint
- <request, topic or pasted plan>
03 SKILL.md 正文
Phase plan
Context: !sh "${CLAUDE_PLUGIN_ROOT}/scripts/py" scripts/guarda.py info
Request: $ARGUMENTS
The JSON above is the context. Write to the user in its language (final answer short, reply_style honored). Use document names from names (keys such as goal, research, progress, plan_review, later, open_questions, need_you, part_closed), never literal. Models and effort per role come from models; if a role is missing use the default in the table. If info failed, say so and use the defaults.
You (main session, highest effort) coordinate: think, decide, dispatch, check. You do not write the long document or map the code.
Roles
| Role | Agent | Default model | Effort |
|---|---|---|---|
| Map code | tune-up:mapper |
haiku | medium |
| Official external docs | tune-up:researcher |
sonnet | medium |
| Write documents | tune-up:writer |
sonnet | medium |
| Pick skills per item | tune-up:skill-finder |
haiku | medium |
| Review the plan | tune-up:plan-reviewer |
opus | high |
- Pass
modelandefforton every call. Never use a model listed inforbidden_subagent_models(default["fable"]). - Independent agents go in ONE message, in background. Do not wait awake; do not read the agent's output file.
- Production: read only. No commit, push or external write unless asked.
- Planning changes no app file. A finding during planning becomes a GOAL item, never a PR.
- If the user said "research only" / "change nothing": stop after Step 2 and use
/tune-up:research.
Workflow scripts
If you orchestrate with the Workflow tool, every agent() carries a literal agentType from the table above (tune-up:mapper, tune-up:researcher, tune-up:writer, tune-up:plan-reviewer), plus model and effort.
A generic agent (no agentType) only for work that needs MCP tools, a browser or images, marked with // tune-up: allow-generic <reason> on the line above.
Template and review-loop rules: see "Workflow scripts" in the part skill.
Step 1 - Prepare
- Read the project profile (
profile_pathfrominfo; ask nothing) and the repo AGENTS/CLAUDE file: plans folder, production, pitfalls. git fetch; note the hash oforigin/<main>. Do not trust a previous session's summary.- Read-only worktree of
origin/<main>in the scratchpad (git worktree add --detach <scratchpad>/wt-main origin/<main>). Every map uses this base and cites the hash. - Number the phase:
ls <plans_dir>/phases | sort -t- -k1,1n | tail -5right BEFOREmkdir(parallel sessions collide). FolderNNN-short-kebab/. - Pasted request: save a verbatim copy as
ORIGINAL-REQUEST.md. Outside plans are adapted to this format, never executed as they came. - If the profile mandates synced folders, run
chmod u+w(macOS/Linux; on Windowsattrib -r) in the SAME command asmkdir. - Small idea (one behavior, few files): plan ONE part with few items; skip extra mappers and the second review round.
Step 2 - Research (one message, all parallel)
- 2 to 4
tune-up:mapper, one theme each. Each request has: base (worktree + hash), numbered questions, "answer with file:line". - 1
tune-up:researcherif there is an external dependency: numbered questions plus the "decisive question" (the one that can kill the plan). Ask for URL, date, status. - You: read-only production queries to measure what the plan assumes (how many, who, since when). A measured number beats a guess.
- Collect everything in scratchpad files (
map-*.md,research-*.md).
Step 3 - Coordinator brief
Write <scratchpad>/<phase>/ brief (name names.brief) from ${CLAUDE_PLUGIN_ROOT}/templates/phase/brief.md. This is where you think: what research overturned, the design, items with size (S/M/L) and executor, criteria C as query/command, Parts, open questions with a recommendation. Be specific (file:line, column names): the writer decides nothing.
Step 4 - Skills per item
For each item call tune-up:skill-finder (one message, parallel) with the item text and the absolute catalog path (info.catalog_path); it returns up to 3 skills (name, SKILL.md path, why). Fill "Skills" in the goal document and "Skills to read" in each work order. No catalog (catalog_path empty or the agent answers none): leave them empty and say so.
Step 5 - Write
One tune-up:writer writes from the brief + maps, each file named by names and shaped by the template in ${CLAUDE_PLUGIN_ROOT}/templates/phase/: goal.md -> names.goal; research.md -> names.research (suffix the date); progress.md -> names.progress; later.md -> names.later. It returns ambiguities: resolve all in the brief and apply in one round.
Step 6 - Plan review (Opus, high)
- Round 1:
tune-up:plan-reviewerwith the base, the paths of goal/research and your main worry. Save asnames.plan_review(templateplan-review.md). - For each finding decide the fix in
<scratchpad>/<phase>/fixes-round1.md; verify doubtful proofs yourself. - The writer applies all and records a "Plan review" section in the goal document.
- Round 2: reviewer checks ONLY that each CRITICAL/HIGH closed, plus new ones. New MEDIUM: you apply. Max 2 rounds; CRITICAL/HIGH left after round 2 ->
names.need_youwith the exact ask.
Step 7 - Owner decisions
AskUserQuestion, up to 4 questions only the owner can decide; recommended first, "(Recommended)" and why in one line. Nothing the code can answer.- Answers go under the
names.decisions_headingheading in goal and progress.names.open_questionskeeps only what stays open. - A decision that adds an item or changes the design: quick map + measurement -> brief -> writer -> focused review of that item only.
Step 8 - Final check
Reread risky items yourself (database, trigger, access rules, external sending). Look for the pitfalls in the profile, plus the pitfall skills of the project's modules (skills whose name or description says pitfalls/armadilhas), plus: a gate that does not test what it decides, non-atomic writes, data already present at deploy, off-then-on, supposed numbers, external service without dedupe. Fix with Edit, confirm with grep.
Step 9 - Close
- Progress document: state, decisions, next owner action,
/goalcommands for ready Parts, each ending with- follow /tune-up:part. Stop when: "<names.part_closed>" in progress OR you wrote <names.need_you> OR 40 turns. - If the project has a Claude memory, record a short note there (facts that change everything, Parts, decisions).
- Short plain answer: the finding that changed the plan, what is decided, what depends on whom, the Part commands. Say nothing was committed.
04 同仓库更多 Skills
instalar
One-step setup of the WhatsApp sales screen on this computer: Docker, the local Evolution API and the sales service, step by step in plain words. Use on: install, set up WhatsApp, prepare sales; instalar, preparar, preparar o WhatsApp, setup WhatsApp, configurar vendas.
video
Makes a hook-first, business-oriented silent vertical reel (9:16, 1080x1920 MP4, 10 or 15 s) from a briefing, using only the owner-approved brand: strategy, the reel composition of the studio kit filled with the script and a hero drawn for this briefing, motion by the studio JavaScript engine (seek-safe, no network), render or slide reserve, and the studio QA. Runs unattended from the panel actions estudio.video and estudio.video-motion. Use on: vídeo, reel, reels, story em vídeo.
post
Makes a hook-first, business-oriented still post or carousel (PNG per frame) from a briefing, using only the owner-approved brand: strategy first, one chosen hook, a composition of the studio kit with a hero drawn for this briefing, one call to action, and every frame checked by the studio QA. Runs unattended from the panel action estudio.post. Use on: post, carrossel, carousel, peça estática, post do Instagram.
marca
Discovers the project's brand (colors, fonts, logo, name) from its own files and shows it to the owner for approval in the panel. Use on: discover the brand, descobrir marca, identidade visual do projeto, marca do estúdio.
marca-ler
Reads the project's brand (marca.json) and says whether the owner approved it. Read-only: never approves, edits or writes anything. Use on: read the brand, is the brand approved, ler marca, marca aprovada.
anuncio
Makes a hook-first, business-oriented ad from a briefing, using only the owner-approved brand: one master image plus the aspect-ratio adaptations the platform needs, text variations inside the platform limits, and an import sheet with everything Paused (Google Ads Editor CSV, or an experimental Meta sheet). Publishes nothing and calls no ad API. Runs unattended from the panel action estudio.anuncio. Use on: anúncio, anuncio, ad, Google Ads, Meta Ads.