iterate-pr

Iterate on PR

One-shot workflow: gather review feedback, fix CI failures, verify locally, commit, push, and reply to all threads.

Requires: GitHub CLI (gh) authenticated.

Important: All scripts must be run from the repository root directory (where .git is located), not from the skill directory. Use the full path to the script via ${CLAUDE_SKILL_ROOT}.

If --dry-run is in $ARGUMENTS, print the plan but make no changes (no commits, no pushes, no replies).

Bundled Scripts

scripts/fetch_pr_checks.py

Fetches CI check status and extracts failure snippets from logs.

uv run ${CLAUDE_SKILL_ROOT}/scripts/fetch_pr_checks.py [--pr NUMBER]

Returns JSON:

{
  "pr": {"number": 123, "branch": "feat/foo"},
  "summary": {"total": 5, "passed": 3, "failed": 2, "pending": 0},
  "checks": [
    {"name": "tests", "status": "fail", "log_snippet": "...", "run_id": 123},
    {"name": "lint", "status": "pass"}
  ]
}

scripts/fetch_pr_feedback.py

Fetches and categorizes PR review feedback using the LOGAF scale.

uv run ${CLAUDE_SKILL_ROOT}/scripts/fetch_pr_feedback.py [--pr NUMBER]

Returns JSON with feedback categorized as:

• high - Must address before merge (h:, blocker, changes requested)
• medium - Should address (m:, standard feedback)
• low - Optional (l:, nit, style, suggestion)
• bot - Informational automated comments (Codecov, Dependabot, etc.)
• resolved - Already resolved threads

Review bot feedback (from Copilot, Claude, CodeQL, etc.) appears in high/medium/low with review_bot: true — it is NOT placed in the bot bucket.

Each feedback item may include:

• thread_id - GraphQL node ID for inline review comments (used for replies)

scripts/reply_to_thread.py

Replies to PR review threads. Batches multiple replies into a single GraphQL call.

uv run ${CLAUDE_SKILL_ROOT}/scripts/reply_to_thread.py THREAD_ID "body" [THREAD_ID "body" ...]

Workflow

1. Identify PR

gh pr view --json number,url,headRefName
gh repo view --json owner,name -q '"\\(.owner.login)/\\(.name)"'

Store PR number and {owner}/{repo} for use in later steps. Stop if no PR exists for the current branch.

2. Gather and Address Review Feedback

Run uv run ${CLAUDE_SKILL_ROOT}/scripts/fetch_pr_feedback.py to get categorized feedback.

Auto-fix (no prompt):

• high - must address (blockers, security, changes requested)
• medium - should address (standard feedback)

When fixing feedback:

• Understand the root cause, not just the surface symptom
• Check for similar issues in nearby code or related files
• Fix all instances, not just the one mentioned

This includes review bot feedback (items with review_bot: true). Treat it the same as human feedback:

• Real issue found → fix it
• False positive → skip, but note why (will be included in reply)
• Never silently ignore review bot feedback — always verify the finding

Prompt user for selection (using AskUserQuestion tool):

• low - use the AskUserQuestion tool to present options. Build the question text with a numbered list of all low-priority items, then offer these options:
  • "All" — address every low-priority item
  • "None" — decline all
  • "Pick…" — let the user specify which numbers in the "Other" text field (e.g., "1,3")

Example question text:

Found 3 low-priority suggestions:
1. [l] "Consider renaming this variable" - @reviewer in api.py:42
2. [nit] "Could use a list comprehension" - @reviewer in utils.py:18
3. [style] "Add a docstring" - @reviewer in models.py:55

Which would you like to address?

Skip silently:

• resolved threads
• bot comments (informational only)

Track every thread's disposition (fixed, declined, false-positive, skipped) for the reply step.

If --dry-run: print the classification table (including the low-priority numbered list for preview, but do NOT prompt the user for selection). Skip to the Summary step.

3. Check CI and Fix Failures

Run uv run ${CLAUDE_SKILL_ROOT}/scripts/fetch_pr_checks.py to get structured failure data.

For each failure:

1. Read the log_snippet and trace backwards from the error to understand WHY it failed
2. Run the failing test(s) locally first. Derive test names from the failure output. If they pass locally, do NOT attempt to fix the code — the failure is likely environmental. Investigate: cache not cleared between tests, transaction.on_commit() callbacks in rolled-back test transactions, test ordering dependencies, or shared mutable state. Report findings to the user.
3. If tests fail locally too, read the relevant code and check for related issues
4. Fix the root cause with minimal, targeted changes

Do NOT assume what failed based on check name alone — always read the logs.

4. Verify Locally

Before committing, verify fixes locally:

• If you fixed a test failure: re-run that specific test
• If you fixed a lint/type error: re-run the linter or type checker on affected files
• For any code fix: run existing tests covering the changed code

Detect the test runner from manifest files (pyproject.toml/setup.py → Python, package.json → Node/TS). Check .github/workflows/ for CI commands first.

If local verification fails, fix before proceeding — do not push known-broken code.

5. Commit and Push

git add <changed files>
git commit -m "fix: address PR feedback and CI failures

- <bullet per change: what was changed and why>"
git push

Store the resulting commit SHA for replies.

6. Reply to All Threads

Inline review threads (items with thread_id): reply using the batch script.

Issue-level comments (items without thread_id): reply via REST API:

gh api -X POST repos/{owner}/{repo}/issues/{pr_number}/comments -f body="$REPLY_BODY"

Use ${CLAUDE_SKILL_ROOT}/scripts/reply_to_thread.py for inline threads. Batch all replies into a single call:

uv run ${CLAUDE_SKILL_ROOT}/scripts/reply_to_thread.py \
  PRRT_abc $'Fixed — description of change.\n\n*— Claude Code*' \
  PRRT_def $'Not applicable — reason.\n\n*— Claude Code*'

Reply format by disposition:

• Fixed: "Fixed in <SHA>. <description of change>.\n\n*— Claude Code*"
• Declined (low): "Noted — declined by author. <brief reason if given>.\n\n*— Claude Code*"
• False positive: "Not applicable — <reason>.\n\n*— Claude Code*"

Guards:

• Before replying, check if the thread already has a reply ending with *— Claude Code* to avoid duplicates
• End every reply with \n\n*— Claude Code*
• If the script fails, log and continue — do not block the workflow

7. Summary

Print a final table:

#	File	Comment (truncated)	Action	Commit / Note
1	src/foo.py:42	"Use Optional instead..."	Fixed	abc1234
2	README.md	"Typo in heading"	Fixed	abc1234
3	src/bar.py:10	"Consider splitting..."	Declined	Low priority, user chose "none"
4	CI: tests	pytest failure in test_api	Fixed	abc1234

Fallback

If scripts fail, use gh CLI directly:

• gh pr checks --json name,state,bucket,link
• gh run view <run-id> --log-failed
• gh api repos/{owner}/{repo}/pulls/{number}/comments