本文へ移動
cccskills
無料GitHub で公開

respond-pr-feedback

Respond to review comments on a PR after evaluation and fixes

インストール方法を見る

含まれるファイル(1)

  • SKILL.md6.9 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Respond to PR Feedback

Post replies to review comments after you've evaluated the feedback and made fixes. Resolves conversation threads by default.

Usage

Invoke the respond-pr-feedback skill, optionally passing these flags:

respond-pr-feedback [--pr <number>] [--no-resolve]

Flags:

  • --pr <number> - PR number to target (default: current branch's PR)
  • --no-resolve - Skip thread resolution after posting replies (default: resolve all)

Prerequisites

Run the fetch-pr-feedback skill first to evaluate the feedback and make any necessary fixes.

Hard gates (sequenced)

Advance only after each pass condition is met (objective checks—not assumed completion).

  1. Prerequisite satisfied: You have already run the fetch-pr-feedback skill, evaluated each thread, and applied fixes or chosen an explicit response strategy per thread (no posting blind).
  2. Context loaded: Step 2 commands exit 0; values parsed from gh pr view, gh repo view, and gh api user are non-empty and assigned to $PR_NUMBER, $PR_AUTHOR, $OWNER, $REPO, $CURRENT_USER—never invent owner, repo, or PR number.
  3. Queue decided: Step 3a jq exits 0. If the filtered list is empty, output exactly All review comments have been addressed. and stop (do not call reply or GraphQL resolve APIs).
  4. Reply before resolve: For each target comment, Step 4 gh api .../replies exits 0 before Step 5 attempts resolution for that item (unless --no-resolve).
  5. Resolve with mapping: Step 5 calls resolveReviewThread only when Step 3b provides a threadId for that comment’s id; if there is no mapping, skip resolution for that item without treating it as failure.

Instructions

1. Parse Arguments

Extract flags from $ARGUMENTS:

  • --pr <number> or detect from current branch
  • --no-resolve flag (boolean, default false)

2. Get PR Context

# Get PR info (if --pr not specified, uses current branch)
gh pr view --json number,author --jq '{number, author: .author.login}'

# Get repo owner/name
gh repo view --json owner,name --jq '{owner: .owner.login, name: .name}'

# Get current authenticated user
gh api user --jq '.login'

Store as $PR_NUMBER, $PR_AUTHOR, $OWNER, $REPO, $CURRENT_USER.

Note: $OWNER, $REPO, etc. are placeholders. Substitute actual values from previous steps.

3. Fetch Unreplied Comments and Thread Data

3a. Unreplied Review Comments

Fetch review comments, excluding PR author and current user, filtering to root comments that haven't been replied to.

Write the jq filter to a temp file using a heredoc with single-quoted delimiter (prevents shell escaping issues with !=, regex patterns, and angle brackets):

cat > /tmp/unreplied_comments.jq << 'JQEOF'
add // [] |
# Root comments from reviewers (not replies, not PR author, not current user)
[.[] | select(
  .in_reply_to_id == null and
  .user.login != $pr_author and
  .user.login != $current_user
)] as $roots |
# IDs that current user has already replied to
[.[] | select(.user.login == $current_user) | .in_reply_to_id] as $replied |
# Filter to unreplied only
$roots | map(select(. as $c | $replied | index($c.id) == null)) |
# Dedup: group by path + line + reviewer, pick newest per group
group_by({
  p: .path,
  l: (.line // .original_line),
  u: .user.login
}) |
map(sort_by(.created_at) | last) |
# Output needed fields
map({
  id,
  user: .user.login,
  path,
  line_display: (
    .line as $end | .start_line as $start |
    if $start and $start != $end then "\($start)-\($end)"
    else "\($end // .original_line)" end
  ),
  body
})
JQEOF
gh api --paginate "repos/$OWNER/$REPO/pulls/$PR_NUMBER/comments" | \
  jq -s --arg pr_author "$PR_AUTHOR" --arg current_user "$CURRENT_USER" \
  -f /tmp/unreplied_comments.jq

If no unreplied comments found, output: "All review comments have been addressed." and stop.

3b. Pre-fetch Thread Data

Fetch review thread IDs to enable resolution after posting replies:

gh api graphql -f query="
  query {
    repository(owner: \"$OWNER\", name: \"$REPO\") {
      pullRequest(number: $PR_NUMBER) {
        reviewThreads(first: 100) {
          nodes {
            id
            isResolved
            comments(first: 1) {
              nodes { databaseId }
            }
          }
        }
      }
    }
  }
"

Build a lookup map: comment databaseId → thread id (unresolved threads only). This enables immediate resolution after posting each reply.

4. Generate and Post Replies

For each unreplied comment, determine the appropriate response based on your evaluation:

Evaluation OutcomeResponse
Feedback was incorrect/unfoundedExplain why the current code is correct
Feedback lacked contextExplain the design decision
Feedback was valid and fixed"Fixed in $COMMIT_SHA" or brief description of change
Feedback was valid but won't fixExplain the tradeoff/decision

Tagging guideline: @-tag bot reviewers (e.g., @coderabbitai) to trigger their processing. Do not @-tag human reviewers.

Post reply to each comment:

gh api "repos/$OWNER/$REPO/pulls/$PR_NUMBER/comments/$COMMENT_ID/replies" \
  -X POST --raw-field body="$RESPONSE"

5. Resolve Threads

This step runs by default. Skip only if --no-resolve was passed.

After posting each reply, look up the $THREAD_ID from the step 3b mapping using the comment's $COMMENT_ID:

gh api graphql -f query="
  mutation {
    resolveReviewThread(input: {threadId: \"$THREAD_ID\"}) {
      thread { isResolved }
    }
  }
"
  • If a comment's $COMMENT_ID has a matching thread ID in the lookup, resolve it
  • If no thread ID found (e.g., issue comment rather than review thread), skip resolution for that comment

6. Output Summary

Group by reviewer:

### Reviewer: coderabbitai[bot]

| File:Line | Response Type | Thread |
|-----------|---------------|--------|
| `src/foo.ts:42` | Fixed in `abc1234` | Resolved |
| `src/bar.ts:15` | Explained design | Resolved |

### Reviewer: octocat

| File:Line | Response Type | Thread |
|-----------|---------------|--------|
| `src/baz.ts:7` | Won't fix | Resolved |

Footer:

**Threads resolved: 3/3**

Response Guidelines

  • @-tag bot reviewers to trigger re-processing; do not tag human reviewers
  • Keep responses concise and technical
  • No performative agreement ("Great point!", "You're right!")
  • Reference specific code/design when explaining decisions
  • If fixed: state what changed, no gratitude

Example

# Respond to all reviewers on current PR (resolves threads)
respond-pr-feedback

# Respond on a specific PR
respond-pr-feedback --pr 123

# Respond without resolving threads
respond-pr-feedback --no-resolve

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Use when you need to mine a conversation, session transcript, or design discussion for architectural decisions before writing ADRs. Identifies problem-solution pairs, trade-off debates, technology choices, and explicit "[ADR]" tags. Triggers on "what decisions did we make", "extract decisions from this chat", "find the choices in our discussion", or "summarize architectural decisions". Also useful after long planning sessions to capture decisions that were made implicitly. Does NOT write ADR documents — use adr-writing or write-adr for that.

日本語の概要は準備中です。原文の説明を表示しています。

existential-birds/beagle822026年9月16日 更新

Use when writing or formatting an ADR document using the MADR template, applying Definition of Done (E.C.A.D.R.) criteria, or verifying ADR completeness. Triggers on "write the ADR", "format as MADR", "check ADR quality", "mark gaps in ADR". Also triggers when a decision has been extracted and needs to become a document. Does NOT extract decisions from conversations (use adr-decision-extraction) or orchestrate the full extract-confirm-write workflow (use write-adr).

日本語の概要は準備中です。原文の説明を表示しています。

existential-birds/beagle822026年9月16日 更新

Use when auditing an agent codebase against the 12-Factor Agents methodology, reviewing LLM-powered system architecture, or assessing agentic app compliance. Triggers on "analyze agent architecture", "12-factor audit", "how compliant is this agent", or "evaluate this LLM app". Also applies when comparing frameworks or planning agent improvements. Not for quick checklists — this performs deep per-factor codebase analysis with file-level evidence.

日本語の概要は準備中です。原文の説明を表示しています。

existential-birds/beagle822026年9月16日 更新

Vercel AI Elements for workflow UI components. Use when building chat interfaces, displaying tool execution, showing reasoning/thinking, or creating job queues. Triggers on ai-elements, Queue, Confirmation, Tool, Reasoning, Shimmer, Loader, Message, Conversation, PromptInput.

日本語の概要は準備中です。原文の説明を表示しています。

existential-birds/beagle822026年9月16日 更新

Reviews App Intents code for intent structure, entities, shortcuts, and parameters. Use when reviewing code with import AppIntents, @AppIntent, AppEntity, AppShortcutsProvider, or @Parameter.

日本語の概要は準備中です。原文の説明を表示しています。

existential-birds/beagle822026年9月16日 更新

Use when the user wants a cited, structured read of local documents and project knowledge. Triggers on: "analyze these docs", "scan my project for context", "read the docs folder", "summarize what's in .beagle/concepts/", "extract context from docs/", "what's in this folder", "go read everything in X and tell me what's there". Also invoked programmatically by other beagle skills (prfaq-beagle Ignition, brainstorm-beagle reference points, strategy-interview context grounding) via the companion contract. Does NOT trigger on codebase lookups ("find this function", "search the repo"), web research (use web-research), LLM-as-judge evaluation (use llm-judge), or document editing (use humanize-beagle). Produces a written scan plan, parallel-subagent findings, and a cited synthesis report on disk — never inline prose, never unsourced claims.

日本語の概要は準備中です。原文の説明を表示しています。

existential-birds/beagle822026年9月16日 更新

existential-birds のスキルをすべて見る

このスキルの問題を報告する