一键导入
council-integration
Supreme Council integration with GSD plan/execute and audit workflows; verdict-handling rules (PROCEED/SIMPLIFY/RETHINK/SKIP)
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Supreme Council integration with GSD plan/execute and audit workflows; verdict-handling rules (PROCEED/SIMPLIFY/RETHINK/SKIP)
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
When the user wants help with paid advertising campaigns on Google Ads, Meta (Facebook/Instagram), LinkedIn, Twitter/X, or other ad platforms. Also use when the user mentions 'PPC,' 'paid media,' 'ROAS,' 'CPA,' 'ad campaign,' 'retargeting,' 'audience targeting,' 'Google Ads,' 'Facebook ads,' 'LinkedIn ads,' 'ad budget,' 'cost per click,' 'ad spend,' or 'should I run ads.' Use this for campaign strategy, audience targeting, bidding, and optimization. For bulk ad creative generation and iteration, see ad-creative. For landing page optimization, see cro.
Persistent engineering knowledge vault with semantic search. Save decisions, patterns, debug solutions, and insights as Zettelkasten notes in Obsidian with automatic embedding indexing for semantic retrieval. Use this skill whenever the user says "/memo", "запомни это", "сохрани в базу", "save this", "remember this", "добавь в базу знаний", "log this pattern", or any variation of wanting to persist knowledge. Also trigger for "/memo find", "/memo search", "что я решал по X", "find in vault", "search vault", "найди в базе" to search notes — both keyword and semantic. Trigger for "/memo dedup", "/memo stats", "/memo project", "/memo reindex". This is the bridge between Claude Code sessions and long-term engineering memory across all projects.
Use when starting any new task to select the right GSD mode (fast/quick/plan-phase) and the right model tier (Haiku/Sonnet/Opus) for subagents — prevents 5-10× cost overpay on routine work
Pack the whole repo (or a remote one) into one AI-friendly file via Repomix. Use for architectural reviews, cross-repo handoff, full-codebase audits — not for single-file lookups. Triggers on "pack repo", "feed codebase to AI", "снапшот", "analyze remote repo".
花叔Design(Huashu-Design)——用HTML做高保真原型、交互Demo、幻灯片、动画、设计变体探索+设计方向顾问+专家评审的一体化设计能力。HTML是工具不是媒介,根据任务embody不同专家(UX设计师/动画师/幻灯片设计师/原型师),避免web design tropes。触发词:做原型、设计Demo、交互原型、HTML演示、动画Demo、设计变体、hi-fi设计、UI mockup、prototype、设计探索、做个HTML页面、做个可视化、app原型、iOS原型、移动应用mockup、导出MP4、导出GIF、60fps视频、设计风格、设计方向、设计哲学、配色方案、视觉风格、推荐风格、选个风格、做个好看的、评审、好不好看、review this design、带解说的动画、解说视频、概念解释视频、长视频科普、配音动画、voiceover、narration、TTS+动画、5分钟讲清楚什么是XX。**主干能力**:Junior Designer工作流(先给假设+reasoning+placeholder再迭代)、反AI slop清单、React+Babel最佳实践、Tweaks变体切换、Speaker Notes演示、Starter Components(幻灯片外壳/变体画布/动画引擎/设备边框/解说Stage)、App原型专属守则(默认从Wikimedia/Met/Unsplash取真图、每台iPhone包AppPhone状态管理器可交互、交付前跑Playwright点击测试)、Playwright验证、HTML动画→MP4/GIF视频导出(25fps基础 + 60fps插帧 + palette优化GIF + 6首场景化BGM + 自动fade)、**带解说的长动画pipeline**(豆包TTS生人声+实测时长生timeline.json+NarrationStage驱动画面+ducking混音→交付HTML实播+发布MP4双形态;铁律:整片是一个连续的运动叙事,禁PowerPoint切换)。**需求模糊时的Fallback**:设计方向顾问模式——从5流派×20种设计哲学(Pentagram信息建筑/Field.io运动诗学/Kenya Hara东方极简/Sagmeister实验先锋等)推荐3个差异化方向,展示24个预制showcase(8场景×3风格),并行生成3个视觉Demo让用户选
Use BEFORE coding any new feature, MVP, pricing/billing change, launch, or pivot. Acts as a product validation gate for solo founders — validates target user, JTBD, pain intensity, current alternative, success metric, distribution channel, structural advantage vs competitors, unit economics with SaaS-graveyard gate, cheapest experiment, and top risk. RIGID — do not proceed to technical planning until product context is documented in `.planning/product/<slug>.md` or risk is explicitly accepted. Triggers on "build", "ship", "add feature", "MVP", "launch", "pivot", "pricing", "billing", before /plan, /tdd, /gsd-plan-phase, /gsd-discuss-phase, or when user describes a feature without naming a measurable success metric.
| name | Council Integration |
| description | Supreme Council integration with GSD plan/execute and audit workflows; verdict-handling rules (PROCEED/SIMPLIFY/RETHINK/SKIP) |
Load this skill whenever the user invokes a GSD workflow command with
--council, when running an audit that needs Council confirmation, or when SKIP/RETHINK verdict handling is required.
Council is a hard gate, not a suggestion.
When /council returns SKIP or RETHINK on a plan, halt the workflow and
surface the verdict to the user. Do NOT proceed to plan-checker, executor,
or merge step until the user either revises the plan or explicitly
overrides the verdict.
| Trigger | What it means |
|---|---|
/gsd-plan-phase --council <N> | Run Council between research and plan-checker |
/gsd-execute-phase --council <N> | Run Council before invoking gsd-executor |
/audit ... --council-review | Audit's Phase 5 Council pass (already mandatory in v4.2+) |
| User asks "should I run Council on this plan?" | Yes — Council is cheap with caching |
When the user types /gsd-plan-phase <N> --council (or attaches --council
to any GSD plan invocation), inject a Council review between research and
plan-checker. Read the phase goal from .planning/phases/<N>-*/goal.md
or .planning/phases/<N>-*/PLAN.md frontmatter, then:
brain "<phase goal text + key constraints from RESEARCH.md>"
Read the verdict from the Council report at
.claude/scratchpad/council-report.md:
| Verdict | Action |
|---|---|
PROCEED | Continue to plan-checker as normal |
SIMPLIFY | Surface concerns to user, ask whether to (a) replan with reduced scope, (b) proceed with the existing plan, (c) abort |
RETHINK | Halt. Print the Skeptic + Pragmatist verdicts. Ask the user to revise the plan. Do not invoke plan-checker. |
SKIP | Halt. Print verdicts and ask whether the phase should be cancelled or rescoped. |
If the user revises the plan, re-run brain (the cache will hit on
identical text — re-prompt the user to bypass with --no-cache if the
revisions are subtle).
For high-risk phases (security work, schema migrations, irreversible
changes), the user attaches --council to invocation. Run Council on
the plan content before spawning the executor:
PHASE_DIR=$(ls -d .planning/phases/<N>-*/ | head -1)
brain "$(cat "$PHASE_DIR/PLAN.md")"
Verdict semantics are the same as plan-phase. RETHINK/SKIP halts execute before any code changes happen.
/audit runs Council mandatorily in Phase 5 (since v4.2). The
--council-review flag noted in older PLAN documents is now the default.
The audit report lands at .claude/audits/<type>-<TIMESTAMP>.md with a
placeholder Council slot:
_pending — run /council audit-review_
Replace it by running:
brain --mode audit-review --report .claude/audits/<type>-<timestamp>.md
run_audit_review() rewrites the slot in place with the verdict table
(REAL / FALSE_POSITIVE / NEEDS_MORE_CONTEXT / disputed). Do NOT manually
edit the slot — the rewrite is atomic and signed.
The display block printed by brain follows this shape:
============================================================
📋 SUPREME COUNCIL REPORT
============================================================
🧐 THE SKEPTIC (Gemini ...):
<full Skeptic prose>
VERDICT: PROCEED|SIMPLIFY|RETHINK|SKIP
🔨 THE PRAGMATIST (ChatGPT ...):
<full Pragmatist prose>
VERDICT: PROCEED|SIMPLIFY|RETHINK|SKIP
------------------------------------------------------------
Skeptic: <verdict>
Pragmatist: <verdict>
Final: <verdict> — <one-line reason>
------------------------------------------------------------
✅|💡|🔄|⛔ VERDICT: <final>
============================================================
Final verdict is the more conservative of the two. Always read
the Final: row, not the individual reviewer rows, when deciding to
halt or proceed.
Identical plan text within TTL (default 7 days) replays the cached report. The display starts with:
♻️ [cached <ts>] Returning previous Council report — no API calls.
(use --no-cache to force a fresh run)
When the user says "the verdict is stale" or "I tweaked the plan", either:
brain --no-cache "<plan>" for a one-off bypass.brain clear-cache (via /council clear-cache) to wipe the
whole cache directory.| Symptom | Cause | Fix |
|---|---|---|
Config not found: ~/.claude/council/config.json | Council not installed | Run bash <(curl -sSL .../setup-council.sh) |
Gemini CLI mode selected but 'gemini' is not in PATH | Gemini CLI missing | npm install -g @google/gemini-cli or switch to API mode in config |
OpenAI CLI mode selected but 'codex' is not in PATH | Codex CLI missing | npm install -g @openai/codex or switch to API mode |
No reviewers available — aborting | Both providers unavailable | Configure at least one of Gemini or OpenAI |
| Verdict feels "off" / outdated | Cache hit on stale plan | brain --no-cache "<plan>" or brain clear-cache |
record_usage not logging | COUNCIL_NO_USAGE_LOG=1 set | Unset the env var |
| Cost gate blocks the call | COUNCIL_COST_CONFIRM_THRESHOLD exceeded | Confirm at the prompt or raise the threshold |
Council is best for plan justification and approach validation, not
line-level code review. Use /audit for code review and /council audit-review for confirming audit findings.
.claude/scratchpad/council-report.mdFinal: row shown explicitly, not just individual rows