--- name: respond-to-pr-review description: Use when PR review comments need to be processed and responded to, when the user says "respond to review", "handle PR comments", "address review feedback", or after receiving code review on a pull request --- # Respond to PR Review ## Overview Process all unaddressed PR review comments in parallel. Each comment gets a dedicated subagent that either applies the fix or replies with technical reasoning for disagreement. **Core principle:** Every comment deserves a thoughtful technical evaluation - not performative agreement or blind implementation. ## Process ```dot digraph { rankdir=TB; detect [label="Detect PR from branch" shape=box]; fetch [label="Fetch top-level comments\n(skip replies)" shape=box]; filter [label="Filter to unaddressed\n(no reply from us)" shape=box]; spawn [label="Spawn parallel subagents\n(one per comment)" shape=box]; done [label="Report summary" shape=box]; detect -> fetch -> filter -> spawn -> done; } ``` ### Step 1: Detect PR and Repo ```bash PR_NUMBER=$(gh pr view --json number --jq '.number') REPO=$(gh repo view --json nameWithOwner --jq '.nameWithOwner') ``` ### Step 2: Fetch Top-Level Review Comments ```bash gh api "repos/${REPO}/pulls/${PR_NUMBER}/comments" \ --paginate \ --jq '[.[] | select(.in_reply_to_id == null) | {id, path, line: (.line // .original_line), body, diff_hunk}]' ``` ### Step 3: Filter to Unaddressed Comments A comment is "addressed" if it already has a reply. Check by fetching all comments and looking at `in_reply_to_id` references: ```bash # Get IDs that have been replied to REPLIED_IDS=$(gh api "repos/${REPO}/pulls/${PR_NUMBER}/comments" \ --paginate \ --jq '[.[] | select(.in_reply_to_id != null) | .in_reply_to_id] | unique') ``` Skip any comment whose `id` appears in `REPLIED_IDS`. ### Step 4: Spawn Subagents Use the **Agent tool** with `subagent_type: "general-purpose"` for each unaddressed comment. Launch all subagents **in parallel** (single message with multiple Agent tool calls). #### Subagent Prompt Template For each comment, provide this prompt to the subagent: ~~~ You are processing a PR review comment. Your job: 1. READ the comment below and understand what change is being requested 2. READ the actual file at the referenced path to understand current code 3. EVALUATE: Does this suggestion make technical sense for this codebase? 4. ACT: Either fix the code OR reply explaining why you disagree ## The Review Comment - **File**: {path} - **Line**: {line} - **Comment**: {body} - **Diff context**: ``` {diff_hunk} ``` ## Rules - If the fix makes sense: Apply it using the Edit tool, then reply confirming: ```bash gh api "repos/{REPO}/pulls/{PR_NUMBER}/comments/{COMMENT_ID}/replies" -f body="Fixed." ``` - If you disagree: Reply with concise technical reasoning: ```bash gh api "repos/{REPO}/pulls/{PR_NUMBER}/comments/{COMMENT_ID}/replies" -f body="[your reasoning]" ``` - Do NOT use performative language ("Great point!", "You're right!") - Do NOT implement blindly - verify the suggestion is correct first - Keep replies short and technical - If the comment contains a "Prompt for AI" section in a
block, you can use it as guidance but still verify independently ~~~ ### Step 5: Report Summary After all subagents complete, report: - How many comments were processed - How many were fixed vs disagreed with - Any that need human attention ## When NOT to Use - Comments from your human partner (handle directly, don't spawn subagents) - Draft PRs with placeholder comments - Comments that are just questions/discussions (not actionable review feedback) ## Common Mistakes | Mistake | Fix | |---------|-----| | Processing reply comments as top-level | Filter `in_reply_to_id == null` | | Responding to already-addressed comments | Check for existing replies first | | Performative agreement in replies | State the fix or technical reasoning only | | Fixing without reading the full file | Always read file before editing | | Ignoring diff_hunk context | Use it to understand what code changed |