Claude/dev profiles task review 011 c uq e vkovei6 rip br f5 ch8 #3
Conversation
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
Adds a comprehensive workflow command for systematically testing, debugging,
and documenting feature implementations with full validation.
This command encapsulates the workflow demonstrated in video streaming fix:
- Phase 1: Test infrastructure setup (Vitest, mocks, comprehensive tests)
- Phase 2: Iterative debugging (logging, diagnostics, Network tab analysis)
- Phase 3: Solution validation (all tests pass, manual verification)
- Phase 4: Comprehensive documentation (session notes, troubleshooting guides)
- Phase 5: Detailed commit strategy
Key principles:
- Tests first to validate fixes work
- Iterate with tests to guide debugging
- Document everything for future sessions
- Simple solutions before complexity
- Comprehensive logging for visibility
- Preserve learnings including failed approaches
Example workflow captured from Emby video streaming fix where:
- Created 41 automated tests (15 unit + 26 integration)
- Debugged 400 errors by checking Network tab
- Fixed API endpoint from /stream.mp4 to /Items/{id}/Download
- Created diagnostic HTML tools
- Documented in session notes and troubleshooting guide
Related to /ta:development-analyst for post-debugging analysis.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Add three essential documentation files for the Task Automation (TA) command system: - ta/README.md: Complete guide to /ta: commands including philosophy, naming conventions, templates, and best practices (383 lines) - ta/commit-and-push.md: Standardized git workflow with testing, detailed commit messages, and PR creation (289 lines) - ta/onboard.md: Project onboarding workflow embodying Compound Engineering philosophy for new Claude sessions (362 lines) Key features: - Compound Engineering philosophy (Kieran Klaassen, Every) - systems that compound in value over time - Clear command structure with phases, examples, safety checks - Proactive contribution mindset for improving workflows - Integration with existing commands and agents These files provide comprehensive documentation for the /ta: command system and enable new Claude sessions to quickly understand and contribute to the workflow library. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
Add automatic CLAUDE.md creation/update to /ta:development-analyst command to close the compound learning loop. Key enhancements: - Automatic CLAUDE.md creation with comprehensive template structure - Smart update strategy that preserves existing content and adds new sessions - No user permission required - runs automatically as part of analysis - Includes project overview, patterns, user preferences, session history - Explains Compound Engineering benefits of automation Technical details: - Reads existing CLAUDE.md first to preserve content - Adds new sessions to top of history (most recent first) - Merges patterns without duplication - Updates Last Updated timestamp - Provides clear summary of what was updated Why this matters: - Closes compound learning loop: analysis → CLAUDE.md → /ta:onboard → next session - Knowledge compounds automatically session-to-session - No manual copy-paste required - Future sessions inherit all accumulated context - Linear progress becomes exponential progress CLAUDE.md is the official Claude Code convention for project memory: - Automatically read by Claude Code on session start - Treated as authoritative system-level instructions - Recursively loaded from current directory upward - Already used in user's other projects (yoke) Files modified: - ta/development-analyst.md - Added 158 lines of CLAUDE.md automation (lines 146-303) This enhancement directly addresses the critical gap identified in our session analysis and embodies the Compound Engineering philosophy. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
Creates new slash command for resetting workspace between sessions by removing untracked artifacts and reverting uncommitted changes. Features: - Assessment phase shows what will be affected - Multiple cleanup options (clean all, selective, stash, cancel) - Safety checks for .env files and large files - Optional build artifact cleanup - Stash support to preserve work in progress Workflow phases: 1. Assessment - Show untracked files and uncommitted changes 2. Cleanup Strategy - Execute based on user preference 3. Verification - Confirm workspace is clean 4. Optional - Clean build artifacts Safety features: - Always shows full list before destructive actions - Warns about .env files (may contain important config) - Warns about large files (>10MB) - Offers stash option to preserve work - Respects .gitignore patterns - Checks for unpushed commits Use cases: - Ending a work session - Starting fresh on new task - Clearing experimental changes - Cleaning up after exploration/debugging Files modified: - ta/clean-session.md - New workflow command (267 lines) - ta/README.md - Added documentation for /ta:clean-session 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
Repositions /ta:clean-session as a defensive, session-start command that guards against context injection, not an end-of-session cleanup. Key philosophy changes: - Run at SESSION START (not end) - Guards against context injection - Ensures only git-tracked content for /ta:onboard - Separation of concerns: NO automatic /ta:development-analyst Workflow: 1. /ta:clean-session (FIRST - defensive hygiene) 2. /ta:onboard (SECOND - load official context) 3. Begin work (clean foundation) Updated messaging: - "Found leftover artifacts from previous session" - "Would pollute context if loaded by /ta:onboard" - "Only git-tracked content remains" - "Ready for onboarding!" Safety checks: - Added: Never run /ta:development-analyst automatically - Clarified: Remove untracked AND uncommitted (not just untracked) - Removed: "Clean if unpushed commits" (was contradictory) Documentation: - New section: "When to Run This Command" - New section: "Session Start Workflow" - Updated: "Separation of Concerns" principles - Removed: Integration with /ta:development-analyst notes Files modified: - ta/clean-session.md - Reframed as session-start hygiene command 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
Integrates /ta:clean-session as Phase 0 of onboarding with 10-second countdown timer. Ensures reliable, clean sessions without requiring user discipline. Phase 0 workflow: 1. Check for untracked files and uncommitted changes 2. Display countdown: "Auto-cleaning in 10 seconds..." 3. User options (10 second window): - [Enter] = Clean now - [s] = Stash artifacts for later - [c] = Cancel (NOT RECOMMENDED) 4. Timeout = Auto-clean (safe default) 5. Confirm clean state before Phase 1 Design philosophy: - No user discipline required (forgetful users protected) - Clean session is priority - Advanced users can stash (10 second window) - Auto-clean is safe default - Context injection prevented by default User experience: - "Where's the any key?" - 10 seconds to react - Advanced users: Press 's' to stash - Forgetful users: Automatic protection via timeout - Clean workspace guaranteed before context loading Benefits: - Reliable: Can't forget to clean - Safe: Auto-clean after timeout - Flexible: Advanced users have options - Clear: Explicit warning about context pollution Files modified: - ta/onboard.md - Added Phase 0 with countdown timer 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
Creates comprehensive session-end workflow that mirrors /ta:onboard, completing the compound engineering loop. Workflow phases: 1. Phase 1: Session Analysis (automatic) - Auto-run /ta:development-analyst - Preserve learnings to CLAUDE.md - Capture patterns, preferences, solutions 2. Phase 2: Commit Work (interactive) - Detect uncommitted changes - Offer commit options (all/tracked/skip/review) - Run /ta:commit-and-push if confirmed - Handle CLAUDE.md inclusion 3. Phase 3: Workspace Cleanup (optional) - Offer to run /ta:clean-session - Remove untracked files - Clean workspace for next session - DRY: Reuses existing command 4. Phase 4: Session Summary - Show what was saved - Confirm all phases completed - Display next steps Design philosophy: - Mirror of /ta:onboard (load ↔ save) - Completes compound engineering loop - DRY: Reuses /ta:development-analyst, /ta:commit-and-push, /ta:clean-session - User has control (interactive choices) - Safe defaults (always analyze, optional cleanup) Session lifecycle: /ta:onboard → Work → /ta:exit → [Next session] → /ta:onboard ↓ ↓ ↓ ↓ Load Build Save Load (compounds!) Special cases handled: - Nothing to commit (skip Phase 2) - CLAUDE.md not committed (warn and offer) - Unpushed commits (offer to push) - Stale untracked files (smart detection) Files modified: - ta/exit.md - New session-end workflow (478 lines) - ta/README.md - Added /ta:exit documentation, updated /ta:clean-session context 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
Implements automatic knowledge preservation system via /ta:development-analyst to enable compound learning across Claude Code sessions. - Initial project memory created with session 2025-10-03 - Documents two-tiered command structure (numbered + TA workflows) - Establishes compound engineering philosophy and patterns - Maps complete system relationships for onboard/exit loops Technical details: - Created by /ta:development-analyst during /ta:exit workflow - Documents 3 established patterns (session lifecycle, symmetric onboard/exit, auto-knowledge preservation) - Catalogues 7 numbered templates + 7 TA workflows + 4 specialized agents - Includes session history, user preferences, and future considerations Testing: - N/A (documentation file, no tests needed) - File structure validated against template - Workflow automation verified (auto-creation works) Files created: - CLAUDE.md - Project memory for compound knowledge across sessions Documentation: - Complete initial knowledge base for future sessions - Enables /ta:onboard to load accumulated context - Closes the compound learning loop (exit saves → onboard loads) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
…tructive default) Problem: - Auto-cleanup in /ta:onboard deleted uncommitted work after 10 second countdown - Lost 5 item-type components + documentation (AlbumTracks, ArtistAlbums, etc.) - Destructive "clean all" was the default auto-execute option Root Cause: - /ta:onboard Phase 0 auto-executed git clean -fd (permanent deletion) - No safe default for forgetful users - Stash option existed but required manual selection Solution: - Changed default auto-execute to STASH (safe, reversible) - Destructive clean now requires explicit user choice (press 'c') - Updated both /ta:onboard and /ta:clean-session to use stash-first philosophy - Added recovery instructions for stashed work Changes: 1. /ta:clean-session.md: - Moved stash to Option A (DEFAULT - SAFEST) - Marked clean all as Option B (DESTRUCTIVE) - Updated user prompt with [s] Stash as default - Auto-execute now stashes instead of cleaning 2. /ta:onboard.md: - Phase 0 countdown now auto-stashes after 10 seconds - Changed messaging: "Auto-stashing" instead of "Auto-cleaning" - Added recovery workflow section with git stash commands - Documented stash-first as implemented feature Impact: - Work is preserved by default (non-destructive) - Clean workspace for onboarding (prevents context injection) - Users can recover stashed work anytime with git stash pop - Destructive operations require explicit confirmation Tested: Not yet - requires real session start to verify countdown behavior Next: Test in next session start, verify stash behavior Prevents: Issue discovered in 2025-10-04 session where item-type components were lost
Problem: - User lost work (5 components + docs) when session ended without commit - /ta:onboard auto-stashed uncommitted work, hiding it - No mechanism to recover stashed work from previous sessions - Users could forget to run /ta:exit entirely Solution: Hybrid approach with two layers of protection Layer 1: /ta:exit (Strong Encouragement) - Phase 2 now shows clear warnings about uncommitted work - Explains consequences: "work will be stashed and hidden" - Offers 5 options (commit all, commit tracked, stash, review, skip) - If user skips: Creates timestamped stash with recovery instructions - No longer just "optional" - strongly recommended with warnings Layer 2: /ta:onboard (Safety Net) - NEW Phase 0.5: Stash Recovery (runs after Phase 0 hygiene) - Detects stashes from previous sessions - Shows contents, age, and file count - Offers: Restore / View / Keep / Discard - Reminds user to commit recovered work Changes: 1. /ta:exit.md: - Updated Phase 2 prompt with clear warnings - Added 5 options instead of 4 (stash is now explicit) - Added "If User Chooses Stash" workflow with instructions - Added "Defense in Depth" section explaining hybrid approach - Added proper/recovery workflow diagrams 2. /ta:onboard.md: - Added Phase 0.5: Stash Recovery (NEW) - Runs between workspace hygiene and context loading - Checks git stash list for previous session work - Interactive recovery with 5 options - Special handling for old stashes (>7 days) - Added complete session lifecycle diagram Philosophy: - Education > Enforcement (teach proper workflow) - Flexibility > Rigidity (allow iteration and exploration) - Safety > Convenience (multiple protection layers) - Recovery > Restriction (make mistakes recoverable) Impact: - Users can skip commit at exit (flexibility) - Work is never lost (safety) - Next session offers recovery (education) - Multiple checkpoints prevent data loss (defense in depth) User Experience: - Proper workflow: /ta:exit → commit → clean start next session - Forgot /ta:exit: Next /ta:onboard detects stash → offers recovery - Skipped commit: Stash created with timestamp → recoverable next session Tested: Not yet - requires real session transitions to verify Next: Test recovery workflow with actual stashed work Prevents: Data loss issue from 2025-10-04 where components were lost Related: Builds on stash-first changes from commit 31c54db
Problem: - Selective file staging (git add file1 file2) requires manual file selection - Risk of forgetting to commit new files Claude creates - Complex conditional logic for which files to include - User wants simple rule: "trust .gitignore, commit everything else" Solution: Use git add -A instead of selective staging Changes: 1. /ta:commit-and-push.md Phase 3: - Added step 1: Verify .gitignore exists and is configured - Changed step 2: Use `git add -A` instead of `git add file1 file2` - Added step 3: Show git status to confirm what will be committed - Updated philosophy section with reasoning 2. Safety Checks section: - Added: Always verify .gitignore exists - Added: Always use git add -A (trust .gitignore) - Added: Show git status after staging - Documented philosophy change with date (2025-10-04) 3. Example session flow: - Updated to show .gitignore check - Shows git add -A instead of selective add - Shows git status --short confirmation 4. /ta:exit.md: - Updated Phase 2 Option 1 to reflect git add -A usage - Added note about 2025-10-04 change - Option 2 now specifies git add -u for tracked-only Philosophy: - OLD: Selective staging with conditional logic - NEW: Trust .gitignore, stage everything - Reasoning: Simpler, more reliable, ensures nothing forgotten - .gitignore is single source of truth Benefits: ✅ New files Claude creates are automatically included ✅ No files forgotten or left behind ✅ Simple, predictable behavior ✅ No conditional logic needed ✅ .gitignore manages exclusions (proper separation of concerns) User Request: "manage proper .gitignore configurations relevant to the project, and add everything else to the repo" Project .gitignore verified (yoke): - Covers: node_modules/, dist/, .env*, *.log, .cache/, coverage/ - Keeps: experiments/, scripts/, docs/ - Well-configured for Vue/Vite project Tested: Not yet - will verify in next commit workflow Next: Use this approach in real commits
Created a comprehensive task review system with 12 professional perspective profiles: NEW FILES (in git repo): - ta/task-review.md - Orchestrator command that solicits feedback from all profiles - Updated ta/README.md - Documented new command and profiles NEW FILES (in ~/.claude/skills/, not in repo): - 12 professional profile skills: * security-engineer.md - Security vulnerabilities, auth, data protection * performance-engineer.md - Performance, scalability, optimization * ux-designer.md - User experience, usability, workflows * devops-engineer.md - Deployment, infrastructure, monitoring * qa-engineer.md - Testing, edge cases, quality assurance * accessibility-specialist.md - WCAG compliance, inclusive design * tech-writer.md - Documentation, clarity, developer experience * product-manager.md - Business value, user needs, requirements * backend-engineer.md - API design, business logic, data modeling * frontend-engineer.md - Components, state management, UI code * database-architect.md - Schema design, queries, data integrity * code-reviewer.md - Code quality, patterns, best practices - skills/README.md - Comprehensive documentation of the profile system KEY FEATURES: - Each profile has clear domain focus and knows when to provide input - Profiles abstain when task isn't relevant to their domain (reduces noise) - Structured feedback with severity levels (Critical, High, Medium, Low) - Consolidated review shows all perspectives in one place - Integration with TodoWrite for creating follow-up tasks - Can review individual tasks or entire todo lists USAGE: /ta:task-review for: Add user authentication endpoint /ta:task-review # (prompts for task from todo list) PHILOSOPHY: - Catch issues before implementation (prevention vs. correction) - Multiple perspectives reduce blind spots - Smart abstention keeps signal-to-noise ratio high - Builds on compound engineering principles NOTES: - Skills files are in ~/.claude/skills/ (user's local Claude config) - Only the orchestrator command (ta/task-review.md) is in this repo - Skills can be invoked individually for deeper discussion - System is extensible - new profiles can be added easily
Added full claude/.claude/ package structure containing all professional perspective profiles and complete documentation. This expands on the initial /ta:task-review orchestrator by including the actual skill files that it invokes. PACKAGE STRUCTURE: claude/.claude/ ├── skills/ # 12 professional perspective profiles │ ├── README.md # Complete skills system documentation │ ├── security-engineer.md │ ├── performance-engineer.md │ ├── ux-designer.md │ ├── devops-engineer.md │ ├── qa-engineer.md │ ├── accessibility-specialist.md │ ├── tech-writer.md │ ├── product-manager.md │ ├── backend-engineer.md │ ├── frontend-engineer.md │ ├── database-architect.md │ └── code-reviewer.md ├── commands/ta/ │ └── task-review.md # Orchestrator (already committed) └── README.md # Package documentation SKILLS FEATURES: - Each profile has distinct professional domain focus - Smart abstention logic (only speak when relevant) - Structured feedback with severity levels - Clear "when to provide input" vs "when to abstain" guidance - Maintains distinct professional voice per role INSTALLATION: Users can copy claude/.claude/ contents to ~/.claude/ to install: - cp -r claude/.claude/* ~/.claude/ Or integrate into dotfiles installer for automatic deployment across environments. USAGE: - /ta:task-review for: [task description] # Multi-perspective review - Skill: [profile-name] # Individual profile consultation This completes the multi-perspective task review system, providing all 12 professional profiles needed for comprehensive pre-implementation feedback.
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.