Menu
Diagnose and fix ai-team GitHub Actions CI failures (lint, test, web-e2e, integration, security). Use when CI is red, the user links an actions run, or before pushing web dashboard / E2E changes.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Runs Ruff, mypy, unit tests, and pip-audit before git push or when preparing for CI/PR. Use --main, --frontend, or --e2e for fuller gates. Use when pushing code, opening a pull request, or when CI failed on ruff, tests, pip-audit, or security.
Run Ruff lint and format before git push or PR. Use when pushing code, when CI failed on Lint, or after editing Python files. Prefer ./scripts/pre_push_check.sh for the full CI gate (includes mypy and pip-audit).
Based on SOC occupation classification
| name | fix-ci |
| description | Diagnose and fix ai-team GitHub Actions CI failures (lint, test, web-e2e, integration, security). Use when CI is red, the user links an actions run, or before pushing web dashboard / E2E changes. |
Mirror .github/workflows/ci.yml. Run from repository root with uv run … locally and in CI.
Single source of truth for unit tests: ./scripts/ci_unit_test.sh (used by CI, pre_push_check.sh, and ci_check.sh).
For local pre-push gates, use the pre-push-checks skill (.cursor/skills/pre-push-checks/).
For pip-audit / bandit only, same skill.
./scripts/ci_check.sh --matrix before push to main (lint + 3.12 cov + 3.11 + security)../scripts/pre_push_check.sh --main if pushing to main.ci.yml)| Step name | 3.11 | 3.12 | Common failure |
|---|---|---|---|
| Run unit tests | no --cov | --cov + fail_under | Real test failure; coverage below threshold |
| Upload coverage artifact | skipped | uploads coverage.xml | Was: missing coverage.xml on 3.11 (fixed with if: matrix.python-version == '3.12') |
| Upload test results | if: always() | same | JUnit artifact even when an earlier step failed |
When 917 passed but job exit 1: if the failing step is Upload coverage artifact on 3.11, the matrix leg ran pytest without --cov — conditional upload fixes it. If the failing step is Run unit tests on 3.12, it is almost always fail_under.
Recent examples:
fail_under on ubuntu-latest.coverage.xml on non-cov leg).Matrix note: fail-fast: false — both Python versions finish; fix the leg that actually failed, not a cancelled sibling from old concurrency.
| GitHub job | Local reproduction |
|---|---|
| Lint | uv run ruff check . → uv run ruff format --check . → uv run mypy src/ |
| Test (3.11) | ./scripts/ci_unit_test.sh --python 3.11 |
| Test (3.12) | ./scripts/ci_unit_test.sh --python 3.12 --cov |
| Both Test legs | ./scripts/ci_check.sh --matrix |
| Web UI E2E | See Web UI E2E below |
| Integration test (main push only) | uv run pytest tests/integration -v --tb=short |
| Security | ./scripts/pip_audit.sh (after pip install --upgrade "pip>=26.1.2") |
With gh authenticated:
gh run view <run-id> --repo RickZee/ai-team --log-failed
gh run view <run-id> --job <job-id> --log-failed
Run in order; stop at first failure and fix before continuing:
uv run ruff check .
uv run ruff format --check .
uv run mypy src/
Fix ruff format (safe auto-fix):
uv run ruff format .
# or one file:
uv run ruff format tests/e2e/web/test_browser_e2e.py
Re-run uv run ruff format --check . until exit code 0.
Fix ruff check: uv run ruff check . --fix where safe; hand-fix the rest.
Fix mypy: type hints / cast / narrow unions in src/ai_team/ only (CI runs mypy src/).
Matches CI web-e2e job:
uv sync
uv run playwright install chromium
cd src/ai_team/ui/web/frontend && npm ci && npm run build
cd - # repo root
uv run pytest tests/e2e/web -m web_e2e -v --tb=short --timeout=120
Faster subsets while iterating:
# API only (no Playwright browser)
uv run pytest tests/e2e/web/test_api_e2e.py -m web_e2e -v --timeout=120
# Browser only
uv run pytest tests/e2e/web/test_browser_e2e.py -m web_e2e -v --timeout=120
Skip frontend build (browser tests skipped):
AI_TEAM_SKIP_FRONTEND_BUILD=1 uv run pytest tests/e2e/web -m web_e2e -v
After the UI restructure, do not use old Dashboard test ids. Grep data-testid in
src/ai_team/ui/web/frontend/src and align tests/e2e/web/test_browser_e2e.py.
| Stale (pre-IA-1) | Current |
|---|---|
nav-dashboard | nav-home |
nav-run (navbar) | home-new-run (Home page CTA → /run) |
nav-artifacts | removed from nav; artifacts on RunDetail tabs |
dashboard-active | run-detail |
dashboard-empty | home-empty |
dashboard-demo | home-demo |
AI-Team Dashboard heading | Runs heading on home-page |
.run-list-assignment | .run-list-description |
Demo lands on / dashboard | Demo lands on /runs/{id} (RunDetail) |
Agent activity is on the Activity tab: click run-tab-activity before asserting
Agent timeline / Activity Log.
| Symptom | Likely cause | Fix |
|---|---|---|
Would reformat: tests/e2e/web/... | Lint, not E2E | uv run ruff format tests/e2e/web/ |
Frontend build did not produce .../index.html | npm build failed | cd src/ai_team/ui/web/frontend && npm ci && npm run build |
playwright / browser missing | Chromium not installed | uv run playwright install chromium |
get_by_test_id("dashboard-active") not found | Stale selectors | See IA-1 selector map |
run-detail / COMPLETE timeout | Demo slow on Linux CI | Demo ~30–60s; tests allow 90s; check _run_demo_async in server.py |
get_by_test_id(...) not found | React data-testid ≠ Python selector | Grep data-testid in frontend/src |
get_by_text("CrewAI") etc. | Copy/layout change on Compare/Run | Update test or restore visible labels |
| API demo timeout | /api/demo never reaches complete | Run test_demo_run_reaches_complete_via_rest alone |
UI change checklist (before push):
uv run ruff format --check .npm run build in src/ai_team/ui/web/frontend./scripts/pre_push_check.sh --e2e (or --main on main)Docs: tests/e2e/web/README.md
CI uses inline uv run pytest in ci.yml (canonical). Local mirror: ./scripts/ci_unit_test.sh / ./scripts/ci_check.sh --matrix.
--cov).--cov + COVERAGE_CORE=sysmon + fail_under from pyproject.toml (currently 55).Local reproduction:
./scripts/ci_unit_test.sh --python 3.12 --cov
./scripts/ci_check.sh --matrix # both Python versions
| Symptom | Likely cause | Fix |
|---|---|---|
| 917 passed, job exit 1, junit 0 failures | Coverage fail_under on Linux (3.12) or missing coverage.xml upload (3.11) | See §0 step map; run ./scripts/ci_unit_test.sh --cov |
tests/unit/test_run_demo.py fails on Linux only | run_demo.main() arms SIGALRM | Mock _install_timeout to return False; load script via importlib with a unique module name (see existing tests) |
demos/02_todo_app not found | Archived demo path in tests | Use demos/02_todo_app (tracked) or tmp_path for txt fixtures; never demos/01_hello_world (archived) |
OpenAIError: Missing credentials on Linux CI only | compile_planning_subgraph eagerly built manager LLM for single-worker mode | Lazy-init manager LLM in supervisor branch only (see planning.py); reproduce with env -u OPENAI_API_KEY in fresh Linux clone |
If only one area changed, run the narrowest tests/unit/... path first, then full unit suite.
Runs on push to main, not on every PR branch. Same command as CI:
uv run pytest tests/integration -v --tb=short
uv run python -m pip install --upgrade "pip>=26.1.2"
./scripts/pip_audit.sh
Same ignore list as .github/workflows/ci.yml (single source: scripts/pip_audit.sh). For the full local gate, use ./scripts/pre_push_check.sh.
Before declaring CI fixed locally:
./scripts/ci_check.sh --matrix # lint + both Python test legs + security
./scripts/pre_push_check.sh --main
--main = lint + mypy + unit tests (with coverage) + integration + frontend build + web E2E + pip-audit — closest match to a main branch CI push.
CI annotations about Node 20 on Actions runners are warnings from actions/*@v4/v5; they do not fail the job. Ignore unless upgrading workflow actions.