Skip to content

status-page: target finelog's DataFusion SQL dialect, not DuckDB #5874

status-page: target finelog's DataFusion SQL dialect, not DuckDB

status-page: target finelog's DataFusion SQL dialect, not DuckDB #5874

Workflow file for this run

name: "Ops - Claude"
on:
issue_comment:
types: [created]
pull_request_review_comment:
types: [created]
issues:
types: [opened, assigned]
pull_request_review:
types: [submitted]
jobs:
claude:
# Only run if comment/issue is from a repo writer (OWNER, MEMBER, or COLLABORATOR) and contains @claude
if: |
(github.event_name == 'issue_comment' &&
(github.event.comment.author_association == 'OWNER' ||
github.event.comment.author_association == 'MEMBER' ||
github.event.comment.author_association == 'COLLABORATOR') &&
contains(github.event.comment.body, '@claude')) ||
(github.event_name == 'pull_request_review_comment' &&
(github.event.comment.author_association == 'OWNER' ||
github.event.comment.author_association == 'MEMBER' ||
github.event.comment.author_association == 'COLLABORATOR') &&
contains(github.event.comment.body, '@claude')) ||
(github.event_name == 'pull_request_review' &&
(github.event.review.author_association == 'OWNER' ||
github.event.review.author_association == 'MEMBER' ||
github.event.review.author_association == 'COLLABORATOR') &&
contains(github.event.review.body, '@claude')) ||
(github.event_name == 'issues' &&
(github.event.issue.author_association == 'OWNER' ||
github.event.issue.author_association == 'MEMBER' ||
github.event.issue.author_association == 'COLLABORATOR') &&
(contains(github.event.issue.body, '@claude') || contains(github.event.issue.title, '@claude')))
timeout-minutes: 30
runs-on: ubuntu-latest
permissions:
contents: write
pull-requests: write
issues: write
id-token: write
actions: read
steps:
- name: Checkout repository
uses: actions/checkout@v5
with:
fetch-depth: 0
- name: Set up Python
uses: actions/setup-python@v6
with:
python-version: '3.12'
- name: Install uv
uses: astral-sh/setup-uv@v5
- name: Run Claude Code
id: claude
uses: anthropics/claude-code-action@fbda2eb1bdc90d319b8d853f5deb53bca199a7c1 # v1.0.140
with:
claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN || secrets.CLAUDE_MAX_OAUTH_TOKEN }}
allowed_bots: 'claude[bot]'
# Empty attribution hides Claude Code's default "Co-Authored-By: Claude"
# / "Generated with Claude Code" trailers from commits and PR descriptions.
settings: '{"attribution":{"commit":"","pr":""}}'
# This is an optional setting that allows Claude to read CI results on PRs
additional_permissions: |
actions: read
claude_args: |
--model claude-opus-4-8
--max-turns 250
--allowedTools "Bash(git:*),Bash(./infra/pre-commit.py:*),Bash(uv:*),Bash(pytest:*),Bash(gh:*),Bash(python:*)"
--system-prompt "Read and follow AGENTS.md (especially its Testing section) and .agents/skills/commit/SKILL.md before starting work. You always have the ability to commit and push your changes. Use your MCP tools if available, falling back to git and gh (which you have authorization for) if needed. NEVER credit yourself: no 'Co-Authored-By: Claude' or 'Generated with Claude Code' trailer in commits, and no self-attribution in PR descriptions. NEVER write tautological, trivial, or slop tests: a test must fail when behavior is wrong, not merely when the code changes; never add a test for a thin wrapper, a one-off script, or just to have one; if a change does not warrant a meaningful test, add none. Before submitting any changes: 1) Run ./infra/pre-commit.py --all-files --fix and fix any issues. 2) Run uv run pytest on tests relevant to your changes. Report on the test status when you provide an update. Always commit and push changes, even if you couldn't get the tests to pass. In any follow-up comments on issues or PRs, state how you tested your changes (which tests you ran and their results)."
triage:
if: |
github.event_name == 'issues' &&
github.event.action == 'opened' &&
(github.event.issue.author_association == 'OWNER' ||
github.event.issue.author_association == 'MEMBER' ||
github.event.issue.author_association == 'COLLABORATOR') &&
!contains(github.event.issue.body, '@claude') &&
!contains(github.event.issue.title, '@claude')
timeout-minutes: 60
runs-on: ubuntu-latest
permissions:
contents: write
pull-requests: write
issues: write
id-token: write
actions: read
steps:
- name: Checkout repository
uses: actions/checkout@v5
with:
fetch-depth: 0
- name: Set up Python
uses: actions/setup-python@v6
with:
python-version: '3.12'
- name: Install uv
uses: astral-sh/setup-uv@v5
- name: Triage Issue
uses: anthropics/claude-code-action@fbda2eb1bdc90d319b8d853f5deb53bca199a7c1 # v1.0.140
with:
claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN || secrets.CLAUDE_MAX_OAUTH_TOKEN }}
allowed_bots: 'claude[bot]'
# Empty attribution hides Claude Code's default "Co-Authored-By: Claude"
# / "Generated with Claude Code" trailers from commits and PR descriptions.
settings: '{"attribution":{"commit":"","pr":""}}'
additional_permissions: |
actions: read
prompt: |
Read AGENTS.md, .agents/skills/fix-issue/SKILL.md, and .agents/skills/commit/SKILL.md.
Then analyze issue #${{ github.event.issue.number }}.
## Actionability check (do this first)
Before doing anything else, read the issue and decide whether it is
actionable. An actionable issue states what is broken or what
needs to be done in concrete terms, names the affected area
(file, module, command, or feature), and (for bugs) gives enough
detail to reproduce or to locate the problem in code. For task
requests, there must be a clear definition of done.
If the issue is too vague, missing reproduction steps, or has no
clear acceptance criteria, do NOT open a PR. Post a
comment listing exactly what additional information you need
(concrete reproduction steps, error messages, expected vs actual
behavior, file/module pointers, or a definition of done) and
stop. Do not guess at intent.
## Comment management (critical)
You must use a SINGLE comment for all your progress updates. Follow this exact pattern:
1. Post your initial plan as a new comment:
`gh issue comment ${{ github.event.issue.number }} --body '🤖 **Plan:** ...'`
2. For ALL subsequent updates, edit that same comment in place:
`gh issue comment ${{ github.event.issue.number }} --edit-last --body '🤖 **Plan:** ...'`
NEVER post a second comment to update progress. Always use --edit-last after the first comment.
Check off completed items and add status notes as you progress.
Use GitHub checkbox syntax for your plan:
```
🤖 **Plan:**
- [x] Completed step
- [ ] Pending step
```
## Respond based on scope
- **Small, self-contained bug fixes (<100 lines changed)**: open a PR following the fix-issue skill workflow.
- **Anything larger** (new features, multi-module changes, design work, research questions, >100 lines):
do NOT open a PR. Instead, update your comment with root cause analysis, relevant code links,
and a proposed approach for a human to review before implementation begins.
## Tests and attribution (critical)
Follow the Testing section of AGENTS.md. NEVER write tautological,
trivial, or slop tests: a test must fail when the behavior is wrong,
not merely when the implementation changes. Do not add a test for a
thin wrapper around a library call, for a one-off script, or just to
have a test. If a change does not warrant a meaningful test, add none.
NEVER credit yourself. Do not add a "Co-Authored-By: Claude" or
"Generated with Claude Code" trailer to commits, and do not
self-attribute in the PR description.
## PR format (critical)
When opening a PR, you MUST follow .agents/skills/commit/SKILL.md exactly.
PR descriptions are plain text — no markdown headers, no bullet lists, no emoji,
no "## Summary" or "## Test plan" sections. Violations will be rejected.
Always produce output — either a PR or an analysis comment. Never silently exit.
claude_args: |
--model claude-opus-4-8
--max-turns 250
--permission-mode acceptEdits
--allowedTools "Bash(git:*),Bash(./infra/pre-commit.py:*),Bash(uv:*),Bash(pytest:*),Bash(gh:*),Bash(python:*)"
--system-prompt "You can commit and push changes. Use MCP tools if available, falling back to git and gh. NEVER credit yourself in commits or PRs (no 'Co-Authored-By: Claude' or 'Generated with Claude Code' trailer). NEVER write tautological or trivial tests — a test must fail when behavior is wrong; do not test thin wrappers or scripts just to have a test. Always commit and push, even if tests fail. Never post multiple comments — post one comment, then use --edit-last for all updates."
autofix:
if: |
github.event_name == 'pull_request_review' &&
(github.event.review.author_association == 'OWNER' ||
github.event.review.author_association == 'MEMBER' ||
github.event.review.author_association == 'COLLABORATOR') &&
contains(github.event.review.body, 'autofix') &&
!contains(github.event.review.body, '@claude')
timeout-minutes: 30
runs-on: ubuntu-latest
permissions:
contents: write
pull-requests: write
issues: write
id-token: write
actions: read
steps:
- name: Checkout PR branch
uses: actions/checkout@v5
with:
fetch-depth: 0
ref: ${{ github.event.pull_request.head.ref }}
- name: Set up Python
uses: actions/setup-python@v6
with:
python-version: '3.12'
- name: Install uv
uses: astral-sh/setup-uv@v5
- name: Apply Autofix
uses: anthropics/claude-code-action@fbda2eb1bdc90d319b8d853f5deb53bca199a7c1 # v1.0.140
with:
claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN || secrets.CLAUDE_MAX_OAUTH_TOKEN }}
allowed_bots: 'claude[bot]'
# Empty attribution hides Claude Code's default "Co-Authored-By: Claude"
# / "Generated with Claude Code" trailers from commits and PR descriptions.
settings: '{"attribution":{"commit":"","pr":""}}'
prompt: |
PR: #${{ github.event.pull_request.number }}
REVIEW: ${{ github.event.review.body }}
A reviewer has requested autofix. Read AGENTS.md (especially its
Testing section) and .agents/skills/commit/SKILL.md.
1. Fetch the full review with inline comments: `gh pr view ${{ github.event.pull_request.number }} --json reviews,comments`
2. Fetch inline PR review comments: `gh api repos/${{ github.repository }}/pulls/${{ github.event.pull_request.number }}/comments`
3. Apply ALL actionable changes from the review comments to the code.
4. Run ./infra/pre-commit.py --all-files --fix
5. Run uv run pytest -m 'not slow' on affected tests
6. Commit and push changes to the PR branch. NEVER credit yourself:
no "Co-Authored-By: Claude" or "Generated with Claude Code" trailer.
7. Post a comment summarizing what was changed, prefixed with "🤖 Autofix:"
When a fix needs a regression test, write a meaningful one — never a
tautological, trivial, or slop test. A test must fail when the
behavior is wrong, not merely when the implementation changes; do not
test a thin wrapper or a script just to have a test.
claude_args: |
--model claude-sonnet-4-6
--max-turns 100
--allowedTools "Bash(git:*),Bash(./infra/pre-commit.py:*),Bash(uv:*),Bash(pytest:*),Bash(gh:*),Bash(python:*)"