Repository navigation
Plan skills and agents implementation - #8
Merged
CoderMariusz merged 10 commits intoDec 10, 2025
Merged
Conversation
Skills Infrastructure: - Add REGISTRY.yaml with 10 generic skills indexed - Create skills directory structure (generic/, domain/) - Add 10 initial skills: supabase-rls, supabase-queries, typescript-patterns, testing-tdd-workflow, react-hooks, nextjs-app-router, api-rest-design, code-review-checklist, git-conventional-commits, typescript-zod New Agents: - SKILL-CREATOR: Creates and maintains skills - SKILL-VALIDATOR: Validates skill accuracy and freshness - TEST-WRITER: Writes test code (TDD RED phase) MCP Profiles: - minimal.json: No MCP (default, context-efficient) - backend.json: Supabase MCP for DB operations - full.json: All MCP servers (use sparingly) CI/CD: - Add GitHub Actions workflow for skills validation ORCHESTRATOR Updates: - Add Skills Integration section - Add skill loading rules - Update routing table with new agents - Document MCP profile usage
Changes: - Add 25 remaining generic skills to complete all 35 planned skills - Update REGISTRY.yaml with all 35 skills (total_skills: 35) - Add AGENT-DEEP-REVIEW.md with detailed analysis of all 18 agents Skills added (25): - API: api-authentication, api-error-handling, api-validation - DevOps: ci-github-actions, docker-basics, env-configuration - Documentation: documentation-patterns, refactoring-patterns - Git: git-workflow - Next.js: nextjs-api-routes, nextjs-data-fetching - React: react-forms, react-performance, react-state-management - Supabase: supabase-auth, supabase-edge-functions, supabase-realtime, supabase-storage - Tailwind: tailwind-patterns - Testing: testing-jest, testing-msw, testing-playwright, testing-react-testing-lib - TypeScript: typescript-api-types, typescript-generics Deep Review findings: - Total agent tokens: ~46,000 - Potential reduction: ~27% (~12,500 tokens) - Identified duplicates: INVEST criteria, accessibility checklist - Recommended 8 new skills to extract from agents Co-Authored-By: Claude <noreply@anthropic.com>
Analysis of existing infrastructure: - Templates (31 files) = STRUCTURE for documents - Patterns (11 files) = PROCESSES for tasks - Checklists (5 files) = VERIFICATION before handoff - Skills (35 files) = KNOWLEDGE for domain expertise Code review findings: - TEST-WRITER: APPROVED ✅ (best structured) - SKILL-CREATOR: Needs Project/Domain skill types - SKILL-VALIDATOR: Needs Project Onboarding workflow Key discoveries: - accessibility.md checklist should become skill - security-backend.md checklist should become skill - UI-PATTERNS.md (496 lines) should become skill Recommendations: - Max 4 skills per task (3% context) - Smaller tasks with 2-3 skills = better quality - Skills loaded ON DEMAND, not all at once Co-Authored-By: Claude <noreply@anthropic.com>
High Priority skills (from existing checklists/patterns): - accessibility-checklist: A11y patterns from WCAG/MDN - security-backend-checklist: OWASP security patterns - ui-ux-patterns: UI patterns for states, layouts, feedback Medium Priority skills (extracted from agents): - invest-stories: INVEST criteria for user stories - discovery-interview-patterns: Requirements gathering patterns - prd-structure: PRD structure with MoSCoW - architecture-adr: ADR format for decisions REGISTRY.yaml updates: - Total skills: 35 → 42 - Added max_skills_per_task: 3 - New categories: UX & Security, Planning & Process Co-Authored-By: Claude <noreply@anthropic.com>
New skills for SKILL-CREATOR and SKILL-VALIDATOR: - research-source-evaluation: Source tiers, search strategies, validation - version-changelog-patterns: Version checking, breaking changes detection - skill-quality-standards: Size limits, confidence levels, required sections SKILL-CREATOR updates: - Added skills.required: [skill-quality-standards] - Added skills.optional: [research-source-evaluation, version-changelog-patterns] - Added Skill Types section (generic/domain/project) - Added Error Recovery section - Added YAML handoff format to VALIDATOR - Reduced from ~550 to ~400 tokens (-27%) SKILL-VALIDATOR updates: - Added skills.required: [skill-quality-standards] - Added skills.optional: [research-source-evaluation, version-changelog-patterns] - Added Project Onboarding Mode section - Streamlined workflow with skill references - Reduced from ~630 to ~450 tokens (-29%) REGISTRY.yaml: total_skills: 42 → 45, version: 1.2.0 Co-Authored-By: Claude <noreply@anthropic.com>
Complete analysis and execution plan for slimming all 16 agents: SKILL-CREATOR & SKILL-VALIDATOR Review: - All MUST HAVE elements complete ✅ - Nice to have: batch creation, auto-split (future) Skills Status: - 45 skills created (sufficient for Phase 1-2) - Only 3 missing: requirements-clarity-scoring, qa-bug-reporting, agile-retrospective Slimming Plan (3 phases): - Phase 1: 7 dev/quality agents, -5,500 tokens (skills exist!) - Phase 2: 3 smaller agents, -1,300 tokens - Phase 3: 6 large agents, -8,700 tokens Total reduction: ~46,000 → ~30,500 tokens (-34%) Includes: - Agent slimming template - Example: BACKEND-DEV 1,810 → 400 tokens (-78%) - Priority order for execution Co-Authored-By: Claude <noreply@anthropic.com>
New skills (completing the set): - requirements-clarity-scoring: 1-5 scoring for requirements gathering - qa-bug-reporting: Bug report structure and severity levels - agile-retrospective: Start/Stop/Continue format Phase 1 Agent Slimming (7 agents): - TEST-ENGINEER: 241 → 93 lines (-61%) - BACKEND-DEV: 248 → 99 lines (-60%) - FRONTEND-DEV: 237 → 101 lines (-57%) - SENIOR-DEV: 231 → 104 lines (-55%) - CODE-REVIEWER: 272 → 105 lines (-61%) - TECH-WRITER: 277 → 104 lines (-62%) - DEVOPS-AGENT: 346 → 115 lines (-67%) Key changes per agent: - Added skills.required and skills.optional in frontmatter - Removed <persona> sections (condensed to Identity) - Removed CRITICAL RULES boxes (knowledge in skills) - Removed detailed patterns (exist in skills) - Kept: Workflow, Decision Logic, Output, Handoff REGISTRY.yaml: total_skills: 45 → 48, version: 1.3.0 Co-Authored-By: Claude <noreply@anthropic.com>
New skills (51 total): - nextjs-server-components: RSC patterns, composition - nextjs-middleware: Auth redirects, headers, matcher patterns - nextjs-server-actions: Form handling, Zod validation, useFormState Phase 2 agent slimming: - UX-DESIGNER: 357 → 101 lines (-72%) - QA-AGENT: 253 → 113 lines (-55%) - SCRUM-MASTER: 287 → 124 lines (-57%) Updates: - FRONTEND-DEV: Added 3 new Next.js skills to optional - REGISTRY.yaml: Updated to v1.4.0 with 51 total skills 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
Phase 3 agent slimming results: - DISCOVERY-AGENT: 639 → 135 lines (-79%) - RESEARCH-AGENT: 559 → 133 lines (-76%) - PM-AGENT: 331 → 121 lines (-63%) - PRODUCT-OWNER: 485 → 140 lines (-71%) - ARCHITECT-AGENT: 343 → 152 lines (-56%) - DOC-AUDITOR: 491 → 170 lines (-65%) Total Phase 3: 2848 → 851 lines (-70%) Skills assigned: - DISCOVERY-AGENT: discovery-interview-patterns, requirements-clarity-scoring - RESEARCH-AGENT: research-source-evaluation, version-changelog-patterns - PM-AGENT: prd-structure, invest-stories, requirements-clarity-scoring - PRODUCT-OWNER: invest-stories, qa-bug-reporting - ARCHITECT-AGENT: architecture-adr, invest-stories, api-rest-design - DOC-AUDITOR: documentation-patterns, code-review-checklist 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
… docs ORCHESTRATOR: - 629 → 307 lines (-51%) - Added skills in frontmatter - ENHANCED multi-track parallel execution documentation - Clear example: 4 tracks (A,B,C,D) running independently - Track A can be in QA while Track D still in Impl - Auto-transition rules preserved TEST-WRITER: - 211 → 98 lines (-53%) - Condensed test templates (detailed in skills) All 20 agents now have skills declared. Total agent lines: 2559 (down from ~5500+ before slimming) 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
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
2 participants
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.
No description provided.