Run local verification before changing the installer, hook behavior, goal state, or completion gates.

​
Static checks

npm run check
Covered:
  • node --check extension.mjs
  • node --check lib/goal-core.mjs
  • node --check adapters/vscode-chat/hook-runner.mjs
  • node --check adapters/vscode-chat/mcp-server.mjs
  • node --check scripts/install.mjs
  • shell syntax checks
  • JSON validation for manifests and hook config

​
Test suite

npm test
Covered:
  • session ID sanitization
  • activation prompt trimming
  • slash-command /goal activation detection
  • sensitive value redaction
  • append-safe evidence merging
  • intentional clearing of remaining and blockers
  • refusal of weak completion evidence
  • unresolved discovered issue blocking
  • evidence-backed issue resolutions for renamed, merged, duplicate, superseded, or clearer-worded issues
  • rejection of wildcard, target-only, or unevidenced issue-resolution claims
  • closed terminal blocked-goal handling
  • multi-session and multi-directory state isolation
  • same-directory ambiguity detection
  • same-directory continuation across resumed copies of the same goal
  • compact prompt notes
  • no raw history leakage in summaries
  • hook quiet exit with no goal
  • subagentStart boundary-only context
  • agentStop blocking while a goal remains open
  • agentStop ignoring terminal blocked goals
  • preCompact snapshot side effect
  • drift blocking helper behavior
  • VS Code Chat hook output shape
  • VS Code Chat prompt activation and continuation
  • VS Code Chat stop blocking
  • VS Code Chat persisted drift tracking
  • VS Code Chat installer MCP merge behavior
  • VS Code extension runtime update-state detection

​
Fixture sanity check

cd tests/fixtures/sample-goal-project
npm test
Expected result: fail before an agent fixes it. The failure is intentional. It proves the runtime E2E prompt is asking Copilot to do real work, not recite hardcoded facts.