## Summary
- Include failure root cause (build error, agent crash, or test failure)
in Slack alerts and GitHub issues
- Include blame context: PR author/commit for PR-triggered failures, or
recent commits/releases for scheduled failures
- Helps triage without clicking through to the CI run
## Example Slack message (scheduled failure)
```
Starter smoke test failing: *pydantic-ai*
Agent crash:
ModuleNotFoundError: No module named 'agent'
No starter code changes — likely a floating dependency update or upstream CopilotKit release
Recent CopilotKit releases:
def5678 chore: release 1.56.0
Run: https://github.com/CopilotKit/CopilotKit/actions/runs/12345
```
## Example (PR failure)
```
Starter smoke test failing: *langgraph-python*
Build failure:
ERR_PNPM_RECURSIVE_RUN_FIRST_FAIL @repo/app@0.1.0 build: next build
Triggered by PR #3700 (max): DRAFT: Add intelligence to langgraph python example
Head: abc123def
Run: https://github.com/CopilotKit/CopilotKit/actions/runs/12345
```
## Test plan
- [ ] Trigger a manual workflow_dispatch run and verify Slack message
includes cause
- [ ] Wait for next scheduled run failure to verify blame context
🤖 Generated with [Claude Code](https://claude.com/claude-code)
GitHub Models (Azure inference) responds slower than OpenAI — the 25s
timeout was causing consistent smoke failures. Increased to 50s
(maxDuration 60s). This is a band-aid until aimock proxy is deployed.
## Summary
- Switch all 3 showcase Slack workflows from raw `curl` to
`slackapi/slack-github-action@v2.1.0` for consistency with the starter
E2E workflow
- Gate drift-detection Slack alerts to schedule-only (`failure() &&
github.event_name == 'schedule'`) so PR failures don't spam #oss-alerts
## Files changed
- `.github/workflows/showcase_drift-detection.yml`
- `.github/workflows/showcase_smoke-monitor.yml`
- `.github/workflows/showcase_redirect-report.yml`
## Test plan
- [ ] Smoke monitor still posts to #oss-alerts on failure
- [ ] Drift detection only posts on scheduled runs, not manual/PR
triggers
🤖 Generated with [Claude Code](https://claude.com/claude-code)
## Summary
- Add automated e2e smoke tests for all 12 CopilotKit starter templates
(`npx copilotkit init` starters)
- Tests run inside Docker using aimock as the LLM backend — no API keys
needed
- 4 test levels per starter: health, agent endpoint, chat round-trip, UI
interaction
- CI runs every 6h (catches floating dep breakage), on release, on PR
changes, and manual dispatch
- Scaffold generator updated to produce test files for new integrations
automatically
## What this catches
The exact bug that hit this week: a CopilotKit release floated
dependencies that broke the `langgraph-python` starter. Users hit a
visible error on page load. These tests would have caught it within 6
hours.
## Architecture
Each starter gets a 3-service Docker Compose stack:
- **aimock** — deterministic LLM mock server
(`ghcr.io/copilotkit/aimock:latest`)
- **agent** — runs the same way users do (`uv run langgraph dev`,
`uvicorn`, etc.)
- **app** — standalone Next.js build with agent URL patched for Docker
networking
Playwright runs inside the Docker network (no host port conflicts, full
parallelization in CI).
## Starters covered
langgraph-python, pydantic-ai, crewai-crews, adk, agno, llamaindex,
strands-python, ms-agent-framework-python, langgraph-js,
langgraph-fastapi, ms-agent-framework-dotnet, mastra
All 12 starters pass all 4 tests (48/48).
## Test plan
- [x] All 12 starters build, start, and pass 4 tests locally
- [ ] CI workflow runs successfully on this PR
- [ ] Verify scheduled trigger fires correctly (after merge)
## Spec
https://www.notion.so/33d3aa38185281e4a93cfc1fdba2aa23🤖 Generated with [Claude Code](https://claude.com/claude-code)
Allowlist-based CI lint that extracts model names from docs MDX code
blocks and validates against docs/model-allowlist.json. Catches stale
model references (gpt-5.2 after gpt-5.4 ships) before they reach users.
- Regex extraction from model="...", model: "...", provider/model patterns
- Provider prefix stripping (openai/, anthropic/, google/, etc.)
- 20 unit tests
Max Korp