Menu
Use when the user says 'new session', 'start a session', 'new worktree', 'start fresh', or 'spin up a workspace'.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Use when executing a preflight checklist. Triggers: 'fly', 'launch execution', 'run the checklist', or when given a preflight checklist file path.
Use when the user says '/learn', 'learn this', 'document this', 'capture this', 'remember this for next time', 'update the docs', asks to capture a durable project learning, or after fixes/reviews reveal reusable patterns.
Use when the user asks to merge, land, rebase before landing, resolve landing conflicts, or advance an integration branch such as main, master, or m3.
Use when the user asks to close, end, clean up, tear down, or finish a worktree/session after work is landed.
Use when starting an interactive parent task session, when the user gives feedback about agent behavior, or when the user asks about observations, skill improvements, or observation logs. Skip delegated/non-interactive subagents, review-only workers, verify-only workers, Codex/Claude print-mode reviewers, and sessions that only report back to a parent agent.
Use when the user says 'deep review', 'thorough review', 'full review', 'triple review', or 'ultra review'.
Based on SOC occupation classification
| name | new-session |
| description | Use when the user says 'new session', 'start a session', 'new worktree', 'start fresh', or 'spin up a workspace'. |
| user-invocable | true |
Set up a worktree with everything needed to run the dev server: env files, node_modules symlink, graphify-out snapshot when available, and a unique port that won't conflict with the main repo or other worktrees.
Check if you're already in a worktree.
git rev-parse --show-toplevel
If the path is inside .claude/worktrees/, you're in a worktree (created externally, e.g. by Claude Code desktop). Skip to step 3.
If you're in the main repo, use the EnterWorktree tool to create one. Let it generate a name, or use one the user provides.
Identify the worktree directory (it should now be your CWD).
Run setup.sh from the worktree:
MAIN_REPO=$(git worktree list --porcelain | sed -n '1s/^worktree //p')
bash "$MAIN_REPO/.claude/skills/new-session/setup.sh"
The script copies env files (.env, .env.local, .env.production.local) from the main repo, symlinks node_modules, copies a local graphify-out snapshot when the main repo has generated graph output, picks a random unused port in 3001-9999, and writes .claude/launch.json. It prints the assigned port to stdout.
If the script lives in ~/.claude/skills/new-session/setup.sh instead (because this is a fresh machine without the project copy), invoke that path directly.
Read PROGRESS.md from the main repo to understand current project state.
Report to the user:
http://localhost:<PORT>npm run dev -- --port <PORT>For the rest of this session, whenever you start a dev server or open a URL, use the port from launch.json - never bare npm run dev (defaults to 3000, which belongs to the main repo) and never http://localhost:3000.
Multiple worktrees share the same database. To avoid conflicts, create worktree-specific test data instead of relying on the shared seed.
"WT: <branch-name>").Schema migrations are the exception. If your feature requires prisma db push with breaking changes, check git worktree list and warn the user before pushing.