Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
46 changes: 31 additions & 15 deletions .github/workflows/claude-code-review.yml
Original file line number Diff line number Diff line change
Expand Up @@ -4,8 +4,7 @@
# Reviews each pull request once with a single Claude Sonnet pass and posts inline comments plus a short summary.
# Later pushes are skipped once the summary (marked <!-- claude-review -->) exists.
# Skips drafts, bots, fork PRs (GitHub withholds secrets from them) and Markdown-only changes. Uses the
# CLAUDE_CODE_OAUTH_TOKEN secret (a Claude subscription token from `claude setup-token`), or ANTHROPIC_API_KEY when
# that isn't set. Without either, the job does nothing.
# ANTHROPIC_API_KEY secret (a Claude API key, billed per review). Without it, the job does nothing.
name: Claude Code Review

on:
Expand Down Expand Up @@ -34,11 +33,11 @@ jobs:
contents: read
pull-requests: write
env:
HAS_CREDENTIALS: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN != '' || secrets.ANTHROPIC_API_KEY != '' }}
HAS_CREDENTIALS: ${{ secrets.ANTHROPIC_API_KEY != '' }}
steps:
- name: Skip without credentials
if: env.HAS_CREDENTIALS != 'true'
run: echo "::notice::Add a CLAUDE_CODE_OAUTH_TOKEN (or ANTHROPIC_API_KEY) repository secret to turn on Claude reviews."
run: echo "::notice::Add an ANTHROPIC_API_KEY repository secret to turn on Claude reviews."
- name: Check for an earlier review
id: earlier
if: env.HAS_CREDENTIALS == 'true'
Expand All @@ -57,23 +56,40 @@ jobs:
if: env.HAS_CREDENTIALS == 'true' && steps.earlier.outputs.reviewed != 'true'
with:
fetch-depth: 1
- uses: anthropics/claude-code-action@1d6de8cb0c237e7c15e9e1bdf973826ebae490cc # v1.0.248
# Claude Code itself rather than claude-code-action: with an API key, the action's runs were refused with "Credit
# balance is too low" (400) on their first request, while the same key works for the plain CLI (November 2026).
- name: Review with Claude Code
id: review
if: env.HAS_CREDENTIALS == 'true' && steps.earlier.outputs.reviewed != 'true'
with:
claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
anthropic_api_key: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN == '' && secrets.ANTHROPIC_API_KEY || '' }}
github_token: ${{ secrets.GITHUB_TOKEN }}
prompt: |
env:
ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The description says the review sends ANTHROPIC_WORKSPACE_ID as an anthropic-workspace-id header via ANTHROPIC_CUSTOM_HEADERS, but neither appears in this diff. If the sk-ant-usr-... key still needs that header, the CLI will hit the same billing error. Add the env var here, or update the description to match what the PR does.

GH_TOKEN: ${{ github.token }}
PROMPT: |
You're reviewing pull request #${{ github.event.pull_request.number }} in ${{ github.repository }}. The PR branch is checked out in the working directory.

1. Run `gh pr view ${{ github.event.pull_request.number }}` and `gh pr diff ${{ github.event.pull_request.number }}` to see what changed and why.
2. Read AGENTS.md and CONTRIBUTING.md, if they exist, for the project's rules. Open other files only when you need context to judge a change.
3. Look for real problems: bugs, crashes, regressions, broken edge cases, security issues, private details (keys, tokens, personal names or emails, server addresses, home-folder paths) and clear breaks of the project's rules. Skip style nitpicks and anything a linter or CI would catch.
4. Post each real problem as an inline comment on the line it's about, with mcp__github_inline_comment__create_inline_comment. Say what's wrong and how to fix it in a sentence or two.
4. Post each real problem as an inline comment on the line it's about, with exactly this command (one per problem, LINE being the line number in the new version of the file):
gh api --method POST repos/${{ github.repository }}/pulls/${{ github.event.pull_request.number }}/comments -f commit_id=${{ github.event.pull_request.head.sha }} -f side=RIGHT -f path=PATH -F line=LINE -f body="..."
Say what's wrong and how to fix it in a sentence or two.
5. Finish with one short summary comment using `gh pr comment ${{ github.event.pull_request.number }} --body "..."`. Start the body with the line `<!-- claude-review -->` (it marks the PR as reviewed). If nothing needs fixing, say so in one line.

Keep it short and only flag things worth fixing. Write in Australian English without em dashes.
claude_args: >-
--model claude-sonnet-5-5
--max-turns 25
--allowedTools "mcp__github_inline_comment__create_inline_comment,Bash(gh pr view:*),Bash(gh pr diff:*),Bash(gh pr comment:*),Read,Grep,Glob"
TOOLS: >-
Bash(gh pr view:*),Bash(gh pr diff:*),Bash(gh pr comment ${{ github.event.pull_request.number }}:*),Bash(gh api --method POST repos/${{ github.repository }}/pulls/${{ github.event.pull_request.number }}/comments:*),Read,Grep,Glob
run: |
npm install -g --no-fund --no-audit @anthropic-ai/claude-code@2.1.327 > /dev/null
claude -p "$PROMPT" --model claude-sonnet-5-5 --max-turns 25 --allowedTools "$TOOLS" --output-format json \
> "$RUNNER_TEMP/claude-execution-output.json"
jq -r '"Claude: \(.num_turns) turns, about US$\(.total_cost_usd * 100 | round / 100)"' "$RUNNER_TEMP/claude-execution-output.json"
jq -e '.is_error | not' "$RUNNER_TEMP/claude-execution-output.json" > /dev/null
# A failed review's reason (a revoked or wrong key, no API credit left, an unknown model), from Claude's result,
# which holds no secrets.
- name: Say why the review failed
if: failure() && steps.review.conclusion == 'failure'
run: |
LOG="$RUNNER_TEMP/claude-execution-output.json"
[ -s "$LOG" ] || { echo "::error::Claude left no output to read."; exit 0; }
# --output-format json writes one result object; its result holds the API's message on a failure
jq -r '"::error::Claude said: \(.result // "nothing" | tostring | .[0:500])"' "$LOG"
Loading