v0.0.2-preview — first release under @azure/functions-skills
Pre-release
Pre-release
·
65 commits
to main
since this release
First release under the new package name @azure/functions-skills.
Package rename
This package was renamed from @agent-loom/azure-functions-skills (last released as v0.12.1). The version line restarted at 0.0.1-preview to reflect the namespace change. When the preview suffix is dropped we will bump straight to v1.0.0, so version-sort confusion with the legacy tags will resolve itself naturally.
For historical reference, the final release under the old name is tagged legacy-final-agent-loom.
Notable changes
- feat(doctor): two-tier project diagnostics with AI semantic analysis (#118)
- Tier 1 (always on, < 1s): deterministic built-in checks for
host.json, runtime version, extension bundle, language version,local.settings.json, deprecated settings, connection strings, function bindings, entry point, TypeScript build - Tier 2 (opt-in, ~60–120s): LLM agent semantic analysis (Copilot CLI / Claude Code / Codex) catching missing error handling, blocking I/O, hardcoded secrets, durable-orchestrator non-determinism
--accept-deep-riskrequired for Tier 2 (agent runs with elevated permissions on workspace)- Output formats:
text(default),json,markdown,html(self-contained, XSS-safe) - Auto-installs skill files via
--install-mode local|plugin
- Tier 1 (always on, < 1s): deterministic built-in checks for
- docs: full README + CONTRIBUTING.md restructure for radically intuitive DX
- README: 11.8 KB → 5.3 KB (-55%), one canonical recipe per topic
- New:
docs/cli-reference.md,docs/doctor-guide.md,docs/skills-vs-azure-skills.md,docs/bad-app-fixtures.md,CONTRIBUTING.mdat repo root
- chore: package optimization
- Removed
src/from npm pack (saved 172 KB unpacked) prepackstripslib/*.js.map(saves 141 KB unpacked)- Result: 110 KB packed (-38% from previous 178 KB)
- Removed
Validation
- 256/256 tests passing (
npm run checkcovers lint + typecheck + validate-skills + verify-plugin-payload + tests + build) - 3 rounds of GPT-5.5 rubber-duck reviews; all blocking findings fixed via TDD
- 96% recall on 16 bad-app fixtures with real Copilot CLI in E2E run
Install
npx @azure/functions-skills install --agent ghcp --dir ./my-appSee the README for quick start.
Follow-up issues filed during review
Important: #109, #110, #111, #112, #113, #114, #115, #116, #117
Documentation: #119, #120 (closed by #123), #121, #122