work

# Work - Incremental Implementation Make incremental progress on Issues, leaving clean state for next session. Based on [Anthropic's Coding Agent pattern](https://www.anthropic.com/engineering/effective-harnesses-for-long-running-agents). ## Arguments This skill accepts an optional issue path: ``` /work [issue-path] ``` Examples: - `/work` - Continue work on current issue from Epic progress.md - `/work issues/20251227-task05-config-store` - Work on specific issue ## When to Use - Starting a new coding session - User says "continue", "work on issue", "next feature" - Running AI agents on parallel issues (via worktrees) ## Session Workflow ``` 1. GET BEARINGS → pwd, git worktree list, cat epics/*/progress.md ↓ 2. READ ISSUE → Read issue.md and feature.json from issue directory ↓ 3. VERIFY BASELINE → Run build/tests, fix any broken state ↓ 4. IMPLEMENT → One feature at a time: implement → test → commit ↓ 5. END SESSION → Use /commit skill, update Epic progress.md ``` ## Directory Structure ``` project/ ├── epics/ │ ├── cli/ │ │ ├── epic.md │ │ ├── progress.md # Check here for current work │ │ └── references/ │ └── website/ │ ├── epic.md │ ├── progress.md │ └── references/ └── issues/ ├── 20250115-task01-auth-setup/ │ ├── issue.md # Requirements │ └── feature.json # Acceptance criteria └── ... ``` ## Reading the Issue Each issue directory contains: - `issue.md` - Requirements, technical details, API design - `feature.json` - Testable acceptance criteria with pass/fail status ```bash # Read issue to understand the work cat issues/ISSUE_ID/issue.md cat issues/ISSUE_ID/feature.json ``` ## Core Loop: Implement One Feature ```bash # 1. Implement the feature # 2. Test end-to-end (browser/HTTP/CLI) # 3. Update feature.json: "passes": true # 4. Use /commit skill to commit changes ``` ## Rules **DO:** - Start every session with "Get Bearings" - Read issue.md and feature.json before implementing - Work on ONE feature at a time - Test BEFORE marking `passes: true` - Use `/commit` skill for commits (not manual git commit) - Update Epic's progress.md at session end **DO NOT:** - Skip baseline verification - Mark features passing without testing - Leave half-implemented code - Commit without using the commit skill ## Launch Agent in New Terminal When user says "work on [issue]" with a specific issue, open a new Ghostty terminal and run the agent there: ```bash # Get current directory first PROJ_DIR=$(pwd) # Open Ghostty and run claude agent in that directory open -a Ghostty --args -e zsh -lc "cd $PROJ_DIR && claude --dangerously-skip-permissions" ``` This allows: - Running an autonomous agent in a separate terminal - Continuing work in the current terminal - Agent runs with full permissions for unattended operation ## Parallel Development (Worktrees) Use Git worktrees to run agents on separate issues in parallel: ```bash # Create worktree for agent git worktree add ../project-task02 -b task02 cd ../project-task02 claude --print "Complete task02 per issue.md" # Continue your work in main repo cd ../project ``` See: [worktree-commands.md](references/worktree-commands.md) | [parallel-agents.md](references/parallel-agents.md) ## Handling Issues | Situation | Action | |-----------|--------| | Broken baseline | Fix FIRST, then new work | | Blocked by dependency | Document, switch issue or end session | | Feature fails testing | Keep `passes: false`, fix and re-test | ## Progress File Format Update the Epic's `progress.md`: ```markdown #### [YYYY-MM-DD HH:MM] Session #N - Working on: task02-auth - Completed: F001, F002 - In progress: F003 - Commits: abc123, def456 - Notes: - Brief description of what was done - Any important decisions or changes ```