Fix CI Failures

This skill automatically diagnoses and fixes CI failures for the current branch by fetching GitHub Actions logs, categorizing the failure, applying targeted fixes, and pushing the result.

Usage

/fix-ci # Auto-diagnose and fix /fix-ci --dry-run # Show what would be fixed without applying

Workflow

Step 1: Fetch CI Failure Details

Determine the current branch:

git branch --show-current

Fetch the latest CI run for this branch:

gh run list --branch <current-branch> --limit 1 --json databaseId,conclusion

• If the latest run has "conclusion": "failure" , fetch the failed logs: gh run view <run-id> --log-failed

• If no failed run is found (conclusion is "success" or no runs exist), report "No CI failures found" and stop.

Step 2: Categorize Failure

Parse the failed log output to determine the failure type(s):

• lint: ruff check errors (look for ruff check output, rule codes like E, F, I, UP, B, SIM)

• format: ruff format errors (look for ruff format output, "would reformat" messages)

• type: mypy errors (look for mypy output, type annotation errors like "Incompatible type", "Missing return")

• test: pytest failures (look for pytest output, "FAILED" markers, assertion errors)

• multiple: if more than one category is detected, fix them in order: format -> lint -> type -> test

Step 3: Apply Fixes

Apply targeted fixes based on the failure category:

For format errors:

ruff format synapse/ tests/

For lint errors:

ruff check --fix synapse/ tests/

For type errors:

• Read the specific mypy error messages from the log

• Identify the file and line number for each error

• Read the source code around the error location

• Apply targeted fixes to the type annotations (add return types, fix incompatible types, add missing imports)

For test failures:

• Read the pytest output to identify failing test names and assertion errors

• Read the failing test code and the source code under test

• Determine whether the fix belongs in the source code or the test

• Apply the targeted fix

After each fix category, run the corresponding check locally to verify before proceeding to the next category.

If --dry-run flag is provided: report what would be fixed for each category but do NOT apply any changes, do NOT commit, and do NOT push.

Step 4: Local Verification

Run all relevant checks to verify the fixes:

ruff check synapse/ tests/

ruff format synapse/ tests/ --check

uv run mypy synapse/

If test failures were involved:

pytest

If any check still fails after the initial fix, attempt one more targeted fix for that category (max 1 retry per category). If it still fails after the retry, proceed to the error handling step.

Step 5: Commit and Push

After all checks pass locally:

Stage only modified files (do NOT stage untracked files):

git add -u

Commit with a descriptive message indicating which categories were fixed:

git commit -m "fix: resolve CI failures (<categories>)"

Where <categories> is a comma-separated list of what was fixed (e.g., lint, format or test or format, lint, type, test ).

Push to the current branch:

git push

Report a summary of what was fixed, including:

• Which failure categories were detected

• Which files were modified

• Whether all local checks pass

Step 6: Error Handling

• gh CLI not authenticated: If gh run list fails with an authentication error, report "GitHub CLI is not authenticated. Run gh auth login first." and stop.

• Fixes do not resolve the issue: If local verification still fails after one retry per category, report exactly what remains broken (include the error output) and stop. Do NOT commit partial fixes that leave CI in a broken state.

• Never force-push: Always use git push , never git push --force or git push --force-with-lease .

Examples

Auto-diagnose and fix all CI failures

/fix-ci

Preview what would be fixed without applying changes

/fix-ci --dry-run

Tool Configuration

This project uses the following CI checks (from pyproject.toml):

Ruff

• Target: Python 3.10

• Line length: 88

• Rules: E, F, I, UP, B, SIM

• Ignored: E501, B008

• Excluded: synapse/proto/a2a_pb2*.py

Mypy

• Strict mode with disallow_untyped_defs

• Tests have relaxed rules (ignore_errors)

• Proto files excluded

Pytest

• asyncio_mode: auto

• Run with: pytest or pytest tests/ -v