You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Fixed Copilot test gates keying on GitHub token presence rather than shape, so an ambient classic or fine-grained PAT no longer un-skips live Copilot tests that cannot pass with it. The env-chain lookup now lives in one exported helper shared by the credential resolver and every test gate, so the two cannot drift apart. Also documented that GH_TOKEN and GITHUB_TOKEN are general-purpose GitHub credentials which the resolver will spend on Copilot inference if they hold a supported token — set GITHUB_COPILOT_TOKEN to keep the credential explicit. (#759)
Other Changes
Integration test reliability: the test harness now forwards a supported GitHub Copilot token from the host into the deployed pod, so the documented AI_PROVIDER=copilot npm run test:integration copilot-provider command actually works instead of deploying a server with no credential. Separately, the recommend workflow tests no longer feed an AI-suggested answer back verbatim when it falls outside the option list the AI generated for that same question — a mismatch that failed the Helm workflow test with validation_failed for reasons unrelated to the workflow. (#783)
tests/ is now covered by lint, format, and typecheck. Previously every quality gate stopped at src/ — npm run lint was eslint src/, npm run format was prettier --write src/, and tsconfig.json excluded **/*.test.ts — so for a test-only change the checks a contributor is asked to run passed without ever examining the file that changed. New format:check and typecheck scripts run in CI alongside the linter, and the same gates now cover packages/agentic-tools. (#784)