ln-914-community-responder（ln 914 社区响应者）

Paths: File paths (shared/, references/, ../ln-*) are relative to skills repo root. If not found at CWD, locate this SKILL.md directory and go up one level for repo root. If shared/ is missing, fetch files via WebFetch from https://raw.githubusercontent.com/levnikolaevich/claude-code-skills/master/skills/{path}.

ln-914-community-responder

Type: L3 Worker (standalone) Category: 9XX Community Engagement Responds to unanswered GitHub Discussions and Issues by analyzing the question, searching the codebase for answers, and composing a helpful reply. Supports single-item and batch modes.

Overview

Phase 0: GitHub Discovery

MANDATORY READ: Load shared/references/community_github_discovery.md

Execute the discovery protocol. Extract:

• {owner}/{repo} for URLs and API calls
• repo.id for GraphQL mutations
• maintainer login (authenticated user)
• Discussion category IDs

Load strategy: check docs/community_engagement_strategy.md in target project, fallback to shared/references/community_strategy_template.md. Extract Section 5 (Engagement Metrics) and Section 6 (Tone Guide).

MANDATORY READ: Load shared/references/community_discussion_formatting.md MANDATORY READ: Load response_styles.md MANDATORY READ: Load shared/references/humanizer_checklist.md

Phase 1: Load Items

Single Item Mode

If $ARGUMENTS contains a number (e.g., 42, #42):

# For discussions
gh api graphql -f query='query($owner: String!, $name: String!) {
  repository(owner: $owner, name: $name) {
    discussion(number: {N}) {
      id number title body
      category { name }
      author { login }
      createdAt
      answerChosenAt
      comments(first: 20) {
        totalCount
        nodes { author { login } body createdAt }
      }
    }
  }
}' -f owner="{owner}" -f name="{repo}"

# For issues (if discussion not found)
gh issue view {N} --repo {owner}/{repo} --json number,title,body,author,createdAt,comments,labels

Batch Mode

If $ARGUMENTS is batch:

1. Fetch recent discussions + issues via GraphQL (last 30 days, open, sorted by created DESC)
2. Filter to P1 items: author!= maintainer AND zero maintainer comments
3. Fetch full context for each item (max 10 per batch)

Interactive Mode

If $ARGUMENTS is empty, list recent unanswered items and ask the user which to respond to.

Phase 2: Analyze Context

For each item:

2a. Understand the Question

1. Read the discussion/issue body — identify the core question or problem
2. Read existing comments — check if partially answered or if follow-ups changed the scope
3. Detect item type: Q&A question, bug report, feature request, configuration help, general feedback

2b. Search Codebase for Answer

Based on the question type, search for relevant information:

FOR each item:
  1. Extract keywords from question (function names, skill names, error messages)
  2. Grep codebase for keywords (max 5 searches)
  3. Read relevant files (max 3 files, prioritize SKILL.md and docs/)
  4. Check git log for recent changes related to the topic
  5. If answer found → proceed to Phase 3
  6. If not found → mark as "needs-manual" and suggest the user respond directly

2c. Detect First-Time Poster

gh api graphql -f query='query($owner: String!, $name: String!) {
  repository(owner: $owner, name: $name) {
    discussions(first: 100) {
      nodes { author { login } }
    }
  }
}' -f owner="{owner}" -f name="{repo}"

Count how many discussions the author has created. If 0 previous (this is their first) → flag for Welcome style.

Phase 3: Classify Response Type

MANDATORY READ: Load response_styles.md — use the classification matrix.

Phase 4: Compose Response

Use the selected style template from response_styles.md.

Required Elements (All Styles)

• Thank the author — by name if possible, acknowledge their contribution
• Answer the question — direct, clear, linked to source code/docs
• Invite follow-up — "Let us know if this helps" or similar
• No jargon without context — explain internal terms
• Link to code — at least one link to relevant file in the repo

Batch Mode Composition

In batch mode, compose all responses first, then present ALL for user review before publishing any.

Phase 5: Fact-Check

Before presenting to user, verify every claim:

1. File paths & links — verify each linked file exists: ls {path}
2. Code references — verify mentioned functions/classes exist: grep -r "{name}"
3. Feature descriptions — re-read source file, confirm accuracy
4. Install/usage commands — verify against README.md
5. Humanizer audit -- run the audit protocol from humanizer_checklist.md. If 3+ AI patterns found, rewrite flagged sections.

Gate: If any check fails, fix the response before proceeding.

Phase 6: Review and Publish

Single Item Mode

Present the composed response to the user. Wait for explicit approval before publishing.

Batch Mode

Present ALL responses in a summary table:

### Batch Responses — {N} items

| # | Type | Title | Style | Status |
|---|------|-------|-------|--------|
| {number} | {Discussion/Issue} | {title} | {Technical/Bug/Welcome/...} | Ready |
| {number} | ... | ... | ... | needs-manual |

Then show each response body. User can approve all, approve selectively, or edit individual responses.

Publish Discussion Comment

gh api graphql -f query='
  mutation($discussionId: ID!, $body: String!) {
    addDiscussionComment(input: {
      discussionId: $discussionId,
      body: $body
    }) {
      comment { url }
    }
  }
' -f discussionId="{discussion.id}" -f body="{response body}"

Publish Issue Comment

gh issue comment {number} --repo {owner}/{repo} --body "{response body}"

Report the comment URL(s) to the user.

Rules

• Always require user approval before publishing any response
• Never close discussions or issues — only respond
• Never mark as answered — let the author do it (for Q&A discussions)
• Batch limit: max 10 items per batch (prevent context overload)
• needs-manual items: report to user with GitHub URL + reason, do not attempt response
• Tone: per strategy Section 6 — respectful, helpful, link to code

Definition of Done

• Items loaded (single, batch, or interactive selection)
• Question context analyzed + codebase searched for answers
• Response type classified per response_styles.md
• First-time posters detected and welcomed
• Response(s) composed with links to relevant code/docs
• Fact-checked (file paths, code references, commands verified)
• User approved response(s)
• Published via GraphQL/CLI, comment URL(s) reported
• needs-manual items reported to user with URLs

Version: 1.0.0 Last Updated: 2026-03-14