| name | paude-launch |
| description | Start a paude container session for fire-and-forget agent work. Use when creating a paude session, delegating a task to paude, running work in an isolated container, setting up a paude worktree, or launching pi/claude/gemini in podman. |
| argument-hint | <session-name> [--slug <slug>] [--spec <absolute-path>] [--agent pi|claude|gemini] |
| allowed-tools | Read Write Shell Glob Grep |
Paude Launch
Start a paude session with correct worktree isolation, git sync, task spec, and domain allowlist — so harvest works later.
<essential_principles>
- Always launch from a worktree, never host
main while other sessions may run.
--prompt-file paths must be absolute — cwd changes inside worktrees.--git is required for harvest; without it the container workspace is empty.- Session name (
paude connect <name>) is separate from git branch slug — record both.
</essential_principles>
- Decision gate: `references/decision-gate.md`
- Flags: `references/create-flags.md`
- Worktrees: `rules/git-worktrees.md`
- Task specs: `.agents/skills/paude-spec/SKILL.md`
- Narrative: `docs/ai-engineering/paude-getting-started.md`
1. Decision gate
Read references/decision-gate.md. If paude is not the right tool, say so and stop.
2. Parse arguments
From $ARGUMENTS:
<session-name> — required--slug <slug> — default: ask or derive from spec filename--spec <path> — task spec; if missing, invoke paude-spec first or ask--agent — default pi for this workspace
3. Ensure spec exists
If no spec file, run or point user to paude-spec before continuing.
Spec must include session name, harvest branch slug, and harvest-prep instructions.
4. Create worktree
From host workspace root (on current main):
git fetch origin && git pull origin main
git worktree add worktrees/<slug> -b <slug>
If worktree exists, ask: resume in existing worktree or new slug?
5. Create session
cd worktrees/<slug>
paude create <session-name> --git --yolo --agent <agent> \
--prompt-file <absolute-path-to-spec>
Read references/create-flags.md for domains, fork image, optional flags.
6. Verify
paude status <session-name>
Confirm: running, project path matches worktree.
7. Hand off to user
Report:
- Session name and worktree path
- How to monitor:
paude status, paude wait
- How to finish:
paude-harvest <session-name> --slug <slug> --merge
<success_criteria>
- Session running with
--git sync from correct worktree
- Spec path was absolute
- User knows session name, slug, and harvest command
</success_criteria>
