菜单
Bridge from /slang-maintain workflow steps to concrete slang-mcp tool calls. Read-only except an optional, user-confirmed post of the generated daily report to Slack. Task recipes live in sibling files.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/shader-slang/slang-skills --skill slang-maintainer-tools命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Use whenever output will name a specific person: release notes, changelogs, contributor lists, commit or PR attribution, code comments, issue and review replies, docs credits, or any prose referring to a human. Governs where a name may come from and how to choose pronouns, so stale names from git history and gender guesses from model priors never reach the output.
Surface the open shader-slang PRs needing human attention with a bundled gh-only Python script (scripts/pr_report.py): an assignee-grouped escalation report computed entirely from live GitHub state (read-only; no writes; optional Discord/Slack mentions). Use for PR triage, the reviewer-attention report, stale-PR follow-up, or a scheduled report run.
Reproduces the shader-slang/slang production PR-review bot (anthropics/claude-code-action@v1 + .github/workflows/claude-pr-review.yml) locally. Same claude CLI, same user prompt, same system-prompt append, same model, same six .claude/agents/* subagents, same deepwiki MCP. The inner CLI produces final-review.md; the wrapping workflow optionally posts it back to GitHub as a COMMENT-state review (event=COMMENT only, never APPROVE/REQUEST_CHANGES) when the orchestrator authorizes (via webhook trigger). Mirrors production's pre-cleanup (minimize prior bot reviews, resolve threads) and post-safety-net (dismiss any non-COMMENT bot reviews). Helper scripts: cleanup.sh (pre), post-review.sh (post). Used by the /slang-pr-review workflow.
正在显示 SKILL.md
基于 SOC 职业分类
| name | slang-maintainer-tools |
| license | MIT |
| description | Bridge from /slang-maintain workflow steps to concrete slang-mcp tool calls. Read-only except an optional, user-confirmed post of the generated daily report to Slack. Task recipes live in sibling files. |
| provides | [] |
| argument-hint | [task: daily-report|release-notes|issue-prioritization|review-messages] [time-range: 24h|7d] [output: file|terminal|both] |
| allowed-tools | Read Write Edit Grep Glob AskUserQuestion mcp__slang-mcp__github_list_pull_requests mcp__slang-mcp__github_get_pull_request mcp__slang-mcp__github_get_pull_request_comments mcp__slang-mcp__github_get_pull_request_reviews mcp__slang-mcp__github_list_issues mcp__slang-mcp__github_search_issues mcp__slang-mcp__github_get_issue mcp__slang-mcp__github_get_discussions mcp__slang-mcp__gitlab_list_merge_requests mcp__slang-mcp__gitlab_list_issues mcp__slang-mcp__gitlab_get_file_contents mcp__slang-mcp__discord_read_messages mcp__slang-mcp__slack_get_channel_history mcp__slang-mcp__slack_get_user_profile mcp__slang-mcp__slack_post_message mcp__slang-mcp__slack_reply_to_thread |
Bridges /slang-maintain (WHAT) to concrete slang-mcp calls (HOW). Read-only, except daily-report may optionally post its finished report to Slack after explicit user confirmation.
| Task | Recipe |
|---|---|
daily-report | daily-report.md |
release-notes | release-notes.md |
issue-prioritization | issue-prioritization.md |
review-messages | review-messages.md |
Each recipe is self-contained: data sources, pagination rules, output template.
/slang-maintain step | Where to look |
|---|---|
| Collect | the task's recipe file — "Data Collection" section |
| Synthesize | the task's recipe file — categorization / dedup rules |
| Deliver | the task's recipe file — "Report Structure" or "Output Format" |
Cross-cutting pitfalls (rate limits, pagination, user-id resolution, squash-merge quirks): gotchas.md.
