feat: add harness engineering improvements#35929
Open
tudorpopams wants to merge 18 commits intomicrosoft:masterfrom
Open
feat: add harness engineering improvements#35929tudorpopams wants to merge 18 commits intomicrosoft:masterfrom
tudorpopams wants to merge 18 commits intomicrosoft:masterfrom
Conversation
📊 Bundle size report✅ No changes found |
|
Pull request demo site: URL |
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
tudorpopams
force-pushed
the
feat/copilot-agents
branch
from
b4ca5af to
eefc38b
Compare
tudorpopams
changed the title
tudorpopams
force-pushed
the
feat/copilot-agents
branch
from
eefc38b to
09fa5eb
Compare
…, docs, and skills improvement Adds five custom Copilot agent profiles (.github/agents/*.agent.md) for: - triage: classify, label, and route incoming issues - fix: implement bug fixes and open PRs - devils-advocate: critically review PRs with confidence scoring - docs-groomer: audit documentation for staleness and gaps - skills-improver: analyze agent performance and improve instructions Also adds global copilot-instructions.md and issue templates for periodic docs audits and skills reviews. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
… skills Adds five agentic workflows (.md + compiled .lock.yml): - agent-triage: classifies, labels, and validates incoming issues - agent-fix: implements bug fixes and opens PRs (triggered by agent:fix label) - agent-review: devil's advocate review with confidence scoring for agent PRs - agent-docs-grooming: weekly documentation audit - agent-skills-improvement: weekly analysis of agent PRs to improve instructions Also includes gh-aw init scaffolding (agentic-workflows agent, .gitattributes). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
These are superseded by the agentic workflows (agent-*.md + .lock.yml). Keeps copilot-instructions.md as global context for all Copilot agents. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Merges global agent context (from .github/copilot-instructions.md) into AGENTS.md alongside the existing Nx guidelines. CLAUDE.md is a symlink to AGENTS.md so both Claude Code and Copilot read the same instructions. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Applies progressive disclosure, mechanical enforcement, and feedback loop patterns from OpenAI's harness engineering approach. 1. Progressive disclosure: Slim down copilot.instructions.md (473→50 lines) and AGENTS.md into pointer-only maps. Detailed docs moved to docs/: - docs/architecture/component-patterns.md (v9 hooks, slots, Griffel) - docs/architecture/design-tokens.md (token categories and usage) - docs/architecture/layers.md (package dependency tiers) - docs/workflows/contributing.md (PR checklist, commands) - docs/workflows/testing.md (test types, SSR safety) - docs/team-routing.md (teams, labels, CODEOWNERS) 2. Agent-guiding error messages: Updated all custom ESLint rules to include doc URLs and fix instructions in error messages so agents self-correct. 3. New ESLint rule: @fluentui/no-hardcoded-style-values — flags hardcoded colors and spacing in .styles.ts files, requiring design tokens. 4. Structural validation: check-v9-structure.js validates that component packages follow the canonical file structure. 5. Layer boundary validation: check-layer-boundaries.js validates that component packages (Tier 3) don't depend on other component packages. 6. Quality tracking: docs/quality-grades.md and docs/tech-debt-tracker.md for agents and engineers to track package quality and debt. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Removes no-hardcoded-style-values ESLint rule, check-layer-boundaries.js, and check-v9-structure.js as requested. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Applies findings from "Increasing AI coding success to 99%" article:
- Add "instructions are source of truth, not existing code" as first line
- Add critical rules as negative constraints ("Never X") in always-loaded context
- Add concrete v9 component code template to override training defaults
- Add legacy anti-patterns section warning about v8 code agents will find
- Add exploration guidance for the large monorepo
- Reframe positive guidance as negative constraints for higher compliance
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Remove all GitHub Agentic Workflow files (.github/workflows/agent-*, .github/agents/, .github/aw/) and clean up related references in AGENTS.md, docs, .prettierignore, and .gitattributes. This PR now focuses solely on harness engineering improvements. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add 5 slash command skills under .claude/skills/: - /v9-component — scaffold a new v9 component with all required files - /change — create beachball change files from current diff - /lint-check — run lint, parse errors, and auto-fix common issues - /token-lookup — map hardcoded CSS values to design tokens - /package-info — quick package lookup (path, deps, owner, tests) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add 6 skills under .agents/skills/ with thin .claude/skills/ wrappers: - /v9-component — scaffold a new v9 component - /change — create beachball change files - /lint-check — run lint and auto-fix common issues - /token-lookup — map hardcoded CSS values to design tokens - /package-info — quick package lookup - /visual-test — visually verify components via Storybook + playwright-cli Replace all npx references with yarn across docs, skills, and copilot instructions to match the repo's package manager. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add /review-pr skill that reviews PRs across 8 categories: change file, V9 patterns, dependency layers, SSR safety, testing, API surface, accessibility, and security/quality. Produces a structured report with a 0-100 confidence score and APPROVE/REQUEST_CHANGES/COMMENT recommendation. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- v9-component: use react-component and react-library generators instead of manually creating files, add generators reference table - change: use `yarn change` (repo's beachball script) and `yarn check:change` for verification - package-info: add useful commands section with Nx executors Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
tudorpopams
force-pushed
the
feat/copilot-agents
branch
from
9f19a68 to
79776c7
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Implement harness engineering best practices for AI agent instructions in the Fluent UI repo.
AGENTS.mdas a concise map, detailed docs indocs/CLAUDE.mdsymlinked toAGENTS.mdfor cross-tool compatibility.agents/skills/with thin.claude/skills/wrappersnpxreferences toyarnacross docs, skills, and copilot instructions to match the repo's package managerWhat's included
Harness Engineering (docs/ restructure)
AGENTS.mdslimmed to ~60 lines as a pointer-only map.github/instructions/copilot.instructions.mdreduced from 473 to ~50 linesdocs/architecture/component-patterns.md— v9 hooks, slots, Griffeldocs/architecture/design-tokens.md— token categories and usagedocs/architecture/layers.md— package dependency tiersdocs/workflows/contributing.md— PR checklist and commandsdocs/workflows/testing.md— test types and SSR safetydocs/team-routing.md— teams, labels, CODEOWNERSdocs/quality-grades.md— per-package quality trackingdocs/tech-debt-tracker.md— known debt items for agentsAgent Skills (
.agents/skills/+.claude/skills/)Skills live in
.agents/skills/(tool-agnostic source of truth). Claude Code wrappers in.claude/skills/reference them via@relative path — nosymlinks, Windows-compatible.
v9-component/v9-component Namechange/changelint-check/lint-check [pkg]token-lookup/token-lookup valpackage-info/package-info pkgvisual-test/visual-test Namereview-pr/review-pr #123ESLint Error Message Improvements
Updated 5 existing rules (
ban-context-export,ban-instanceof-html-element,no-global-react,no-restricted-imports,no-context-default-value) toinclude fix instructions and
docs/references in error messages.npx → yarn
All command references in docs, skills, and copilot instructions updated from
npxtoyarnto match the repo's package manager.Test plan
CLAUDE.mdsymlink resolves correctly toAGENTS.md/menu in Claude Code/package-info react-buttonreturns correct package details/token-lookup #0078d4suggests the right design token/visual-test Buttonlaunches Storybook and captures a screenshot/review-prproduces structured output with confidence score