|
| 1 | +# AI agent framework tasks |
| 2 | + |
| 3 | +This checklist tracks repository updates that improve AI workflows using agentic primitives, context engineering, and spec-first development. |
| 4 | + |
| 5 | +## Option 1 — Docs-first primitives (low effort, high ROI) |
| 6 | + |
| 7 | +- [x] Create domain instructions files: |
| 8 | + - [x] .github/instructions/managed.instructions.md |
| 9 | + - [x] .github/instructions/native.instructions.md |
| 10 | + - [x] .github/instructions/installer.instructions.md |
| 11 | + - [x] .github/instructions/testing.instructions.md |
| 12 | + - [x] .github/instructions/build.instructions.md |
| 13 | +- [x] Add role-scoped chat modes with tool boundaries: |
| 14 | + - [x] .github/chatmodes/managed-engineer.chatmode.md |
| 15 | + - [x] .github/chatmodes/native-engineer.chatmode.md |
| 16 | + - [x] .github/chatmodes/installer-engineer.chatmode.md |
| 17 | + - [x] .github/chatmodes/technical-writer.chatmode.md |
| 18 | +- [x] Add context and memory anchors: |
| 19 | + - [x] .github/context/codebase.context.md |
| 20 | + - [x] .github/memory.md |
| 21 | +- [x] Reference these entry points from onboarding: |
| 22 | + - [x] Link instructions, chat modes, and context in .github/copilot-instructions.md |
| 23 | + |
| 24 | +## Option 2 — Agentic workflows + spec-first flow (moderate effort) |
| 25 | + |
| 26 | +- [ ] Prompts in .github/prompts/: |
| 27 | + - [ ] feature-spec.prompt.md (spec → plan → implement with gates; uses spec-kit) |
| 28 | + - [ ] bugfix.prompt.md (triage → RCA → fix plan → patch + tests) |
| 29 | + - [ ] test-failure-debug.prompt.md (parse NUnit output → targeted fixes) |
| 30 | +- [ ] Specification templates: |
| 31 | + - [ ] .github/spec-templates/spec.md and plan.md (or link to spec-kit) |
| 32 | + - [ ] .github/recipes/*.md playbooks for common tasks |
| 33 | +- [ ] Fast inner-loop tasks: |
| 34 | + - [ ] Extend .vscode/tasks.json: quick builds (managed/native), smoke tests, whitespace/gitlint |
| 35 | + |
| 36 | +## Option 3 — Outer-loop automation + MCP integration (higher effort) |
| 37 | + |
| 38 | +- [ ] Copilot CLI/APM scaffolding: |
| 39 | + - [ ] apm.yml: map scripts to prompts and declare MCP dependencies |
| 40 | + - [ ] Document local usage: `apm install`, `apm run copilot-feature-spec --param specFile=...` |
| 41 | + - [ ] GH Action to run chosen prompt on PR, post summary/comments |
| 42 | +- [ ] MCP servers & boundaries: |
| 43 | + - [ ] Add GitHub MCP server and Filesystem MCP (pilot set); restrict by chat mode |
| 44 | + - [ ] Capture list and policies in `.github/context/mcp.servers.md` |
| 45 | +- [ ] CI governance: |
| 46 | + - [ ] lint-docs job to verify COPILOT.md presence/links and src-catalog consistency |
| 47 | + - [ ] prompt validation job to parse `.prompt.md` frontmatter/structure |
| 48 | +- [ ] Security & secrets: |
| 49 | + - [ ] Use least-privilege tokens (e.g., `secrets.COPILOT_CLI_PAT`) |
| 50 | + - [ ] Add a security review checklist for enabling new tools/servers |
| 51 | +- [ ] Rollout strategy: |
| 52 | + - [ ] Pilot a no-write prompt (`test-failure-debug.prompt.md`) on PRs |
| 53 | + - [ ] Iterate then enable selective write-capable workflows |
| 54 | + |
| 55 | +See: `.github/option3-plan.md` for details. |
| 56 | + |
| 57 | +## Notes |
| 58 | +- Keep instructions concise and domain-scoped (use `applyTo` when appropriate). |
| 59 | +- Use consistent headings across COPILOT.md: Responsibilities, Entry points, Dependencies, Tests, Pitfalls, Extension points. |
| 60 | +- Prefer fast inner-loop build/test paths for agents; reserve installer builds for when necessary. |
| 61 | + |
| 62 | + |
| 63 | +## small but high-impact extras |
| 64 | +- [ ] Add mermaid diagrams in .github/docs/architecture.md showing component relationships (Cellar/Common/XCore/xWorks), so agents can parse text-based diagrams. |
| 65 | +- [ ] Create tests.index.md that maps each major component to its test assemblies and common scenarios (fast lookup for agents). |
| 66 | +- [ ] Enrich each COPILOT.md with section headers that match your instructions architecture: Responsibilities, Entry points, Dependencies, Tests, Pitfalls, Extension points. Agents recognize consistent structures quickly. |
| 67 | +- [ ] Link your CI checks in the instructions: we already added commit/whitespace/build rules and a PR template—keep those links at the top of copilot-instructions.md. |
0 commit comments