跳到正文
cccskills
免费在 GitHub 公开

document-release

Update in-repo docs (README, CLAUDE.md, architecture, INDEX) to match what shipped. Use after a merge, project, or release.

中文简介正在准备中。当前显示原文说明。

indigoai-us/hq-core星标 90(GitHub 90 + cccskills 0)

实测评分 未实测

查看安装方法

包含的文件(2)

  • SKILL.md9.3 KB
  • agents/openai.yaml120 B

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Post-Ship Documentation Sync

You are a technical writer performing a post-ship documentation audit. Your job: make every doc reflect what actually shipped — no more, no less.

Critical rules:

  • NEVER clobber CHANGELOG entries — polish wording only, never regenerate or rewrite history
  • NEVER bump VERSION without asking
  • Use Edit tool with exact matches — never use Write to overwrite CHANGELOG or README
  • AUTO-UPDATE machine-consumed docs (CLAUDE.md, architecture, INDEX.md) — factual accuracy matters more than voice
  • ASK before updating user-facing docs (README.md) — voice and framing matter

Codex/Handoff Mode

Codex /handoff runs this skill in a visible subagent, not through core/scripts/handoff-post.sh or claude -p. In that mode:

  • DO NOT call AskUserQuestion — return proposed user-facing edits in the final summary instead
  • AUTO changes apply normally — CLAUDE.md, architecture docs, INDEX.md, prd.json/board.json status flips
  • ASK changes are logged in the subagent final report, not applied — include the file, one-line reason, and a concise before/after or diff summary
  • Scope gate still applies — only run when files_touched includes companies/ or repos/ paths, as provided by the caller

Legacy Headless Mode

If $CLAUDE_HEADLESS=1 (legacy detached launchers only), this skill runs non-interactively:

  • DO NOT call AskUserQuestion — there is no user to answer
  • AUTO changes apply normally — CLAUDE.md, architecture docs, INDEX.md, prd.json/board.json status flips
  • ASK changes are logged, not applied — for each proposed README/setup-guide edit, print a block to stdout:
    PROPOSED ({file}): {one-line summary}
    ---
    {unified diff or before/after snippet}
    ---
    
    stdout is captured by the caller; the user reviews next session.
  • No prompts, no blocking — if a decision would require user input, default to "skip + log" and continue
  • Scope gate still applies — only run when files_touched includes companies/ or repos/ paths (enforced by caller)
  • Exit cleanly — emit the final report to stdout so the log captures the summary

Shell check at the top of the run:

if [[ "${CLAUDE_HEADLESS:-0}" = "1" ]]; then
  echo "[document-release] headless mode — AskUserQuestion disabled, ASK changes will be logged"
fi

Step 0: Company + Project Resolution

Same company anchor pattern as all HQ commands:

  1. Resolve via bash core/scripts/resolve-company.sh --prompt "{input}" (session bind first, then a whole-token scan of the input — not the first word alone)
  2. If matched: set {co}, load policies
  3. Resolve project: find companies/{co}/projects/{slug}/prd.json or infer from recent git activity

If no project slug provided:

  • Check recent git log for project-related commits
  • Check workspace/orchestrator/ for recently completed projects
  • AskUserQuestion if still ambiguous

Announce: Documenting release for: {project title} ({co})

Step 1: Diff Analysis

Determine what changed since the project started:

# If project has a known start commit or date
git log --oneline --since="{prd.createdAt}" -- {repo paths}
git diff {start_ref}..HEAD --stat -- {repo paths}

# If no clear start ref, use the PRD creation date
git log --oneline --since="{prd.metadata.createdAt}" -- {repo paths}

Identify and categorize:

  • New files — may need doc mentions (new components, APIs, tools)
  • Removed files — doc references to remove
  • Renamed/moved files — doc references to update
  • Changed APIs — function signatures, endpoints, env vars
  • New dependencies — package.json/requirements changes
  • New env vars — .env.example or setup doc updates needed
  • Config changes — deployment, CI, infrastructure

Present summary:

Diff analysis:
- Files added: {N}
- Files removed: {N}
- APIs changed: {list}
- New env vars: {list}
- New dependencies: {list}

Step 2: Doc Audit

Scan for all documentation files in scope. For each, classify the update strategy:

FileStrategyRationale
README.mdASKUser-facing, voice matters. Present proposed changes for approval
CLAUDE.mdAUTOMachine-consumed, factual accuracy is critical
.claude/CLAUDE.md (repo-level)AUTOMachine-consumed
Architecture docs (docs/*.md, ARCHITECTURE.md)AUTOMust match code reality
INDEX.md filesAUTOStructural, regenerable
CHANGELOG.mdPOLISH ONLYNever rewrite. Only fix typos, formatting, or broken links in existing entries
prd.json statusAUTOSet to "completed" if all stories done
board.json statusAUTOSet to "completed"
API docs (openapi.yaml, route comments)AUTOMust match implementation
Setup/install guidesASKUser-facing, may affect onboarding

For each doc file found:

  1. Read current content
  2. Cross-reference against the diff (Step 1)
  3. Identify stale references (files/functions/APIs that no longer exist)
  4. Identify missing references (new files/functions/APIs not yet documented)
  5. Classify each needed change as AUTO or ASK

Step 3: Apply Updates

AUTO updates (apply directly)

Use Edit tool with exact string matches. Make minimal, targeted edits:

  • Update file paths that changed
  • Update function/class names that were renamed
  • Update API endpoint documentation to match implementation
  • Remove references to deleted code
  • Add references to new significant code (new modules, services, APIs)
  • Update CLAUDE.md sections that reference project structure
  • Regenerate affected INDEX.md files
  • Set prd.json status → "completed" (if all stories done)
  • Set board.json status → "completed"

ASK updates (present for approval)

Interactive mode (default): For each user-facing doc change, present via AskUserQuestion:

README.md update needed:

Current: {relevant section}
Proposed: {updated section}

A) Apply this change
B) Skip — I'll update manually
C) Modify — let me adjust the wording

Headless mode ($CLAUDE_HEADLESS=1): Do NOT call AskUserQuestion. Log the proposal to stdout and continue:

PROPOSED (README.md): {one-line summary}
---
--- before
{current snippet}
+++ after
{proposed snippet}
---

Counts as "logged, not applied" in the final report. User addresses these next session.

CHANGELOG rules

  • NEVER regenerate CHANGELOG entries
  • NEVER reorder entries
  • Only fix: typos, broken links, formatting inconsistencies
  • If the project should have a CHANGELOG entry and doesn't: ASK the user, don't auto-generate

Step 4: Cross-Doc Consistency Check

After all updates are applied, verify consistency:

  1. Internal links — grep for [...](path) markdown links, verify targets exist
  2. Code references — grep for backtick-quoted function/file names, verify they still exist in the codebase
  3. Env var references — verify all documented env vars exist in .env.example or equivalent
  4. Import paths — verify documented import examples match actual file locations
  5. Stale examples — flag code examples that reference removed or renamed APIs

For each broken reference:

Consistency issue: {file}:{line}
References: {stale reference}
Current: {what it should be, or "removed"}
Action: {AUTO-FIX / FLAG}

Auto-fix simple renames. Flag removals for user review.

Step 5: Cleanup + Status Sync

Project status updates

If all stories in prd.json are done:

  • Set prd.json → metadata.status: "completed", metadata.completedAt: "{ISO8601}"
  • Update board.json entry → status: "completed", updated_at: "{ISO8601}"

Reindex

qmd update 2>/dev/null || true

Final report

Document release complete: {project title}

Updated:
  AUTO: {N} files ({list})
  ASK:  {N} files ({list})
  SKIPPED: {N} files ({list})

Consistency:
  Links verified: {N}
  Issues found: {N} ({N} auto-fixed, {N} flagged)

Status:
  prd.json: {completed/unchanged}
  board.json: {completed/unchanged}

Rules

  • NEVER clobber CHANGELOG — this is the #1 rule. Polish wording, never regenerate
  • NEVER bump VERSION without asking — version bumps are release decisions, not doc decisions
  • Minimal diff — use Edit with exact matches, never Write to overwrite docs wholesale
  • AUTO for machines, ASK for humans — CLAUDE.md and architecture docs can be auto-updated; README and setup guides need approval
  • Company isolation — only touch docs within the resolved company/project scope
  • INDEX.md regeneration — use the standard INDEX.md spec from core/knowledge/public/hq-core/index-md-spec.md
  • No implementation — this command updates documentation only. If a doc change reveals missing code, flag it — don't write the code
  • Do NOT use TodoWrite or EnterPlanMode
  • Idempotent — running this command twice should produce no additional changes
  • Headless-safe — when $CLAUDE_HEADLESS=1, never call AskUserQuestion; log proposals to stdout instead. This skill is invoked headless by core/scripts/handoff-post.sh, and a prompt in that context blocks forever

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

accept

免费

Accept HQ membership invites from a magic link or raw token.

中文简介正在准备中。当前显示原文说明。

indigoai-us/hq-core902026年10月26日 更新

adr

免费

Capture qualifying architecture decisions as ADRs in the right HQ or repo location.

中文简介正在准备中。当前显示原文说明。

indigoai-us/hq-core902026年10月26日 更新

Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction.

中文简介正在准备中。当前显示原文说明。

indigoai-us/hq-core902026年10月26日 更新

architect

免费

Find shallow modules and propose deepening refactors, ranked with file refs; never edits code. Use when code is hard to change or planning a refactor.

中文简介正在准备中。当前显示原文说明。

indigoai-us/hq-core902026年10月26日 更新

Generate dark ASCII banner graphics for HQ and social launch assets.

中文简介正在准备中。当前显示原文说明。

indigoai-us/hq-core902026年10月26日 更新

Compare approaches to best-in-class prior art, merge existing work, verify cruft removal, and recommend the most durable design before PRD work.

中文简介正在准备中。当前显示原文说明。

indigoai-us/hq-core902026年10月26日 更新

查看 indigoai-us 的全部技能

报告此技能的问题