autopilot

name	autopilot
description	Run one fully-autonomous development iteration. Picks a ticket (from --focus or backlog via ideator), then runs the `orchestrator autopilot <slug>` workflow subcommand. This skill should be used when the user says 'autopilot', 'autonomous', 'self-improve'.
user-invocable	true
args	[{"name":"focus","description":"Optional ticket ID, slug, or focus hint. If provided and it looks like a ticket/slug, use it directly; otherwise pass to ideator as a focus area.","required":false}]

Input

$ARGUMENTS

Overview

/autopilot is a thin wrapper that runs one complete autonomous feature workflow.

• If $ARGUMENTS is empty → invoke ideate --next to pick the most valuable backlog item.
• If $ARGUMENTS looks like a ticket/slug (matches ^[A-Z]+-\d+$ or ^[a-z][a-z0-9-]*$) → use it directly.
• Otherwise treat it as a focus hint → invoke ideate --next --focus "<hint>".

Then run orchestrator autopilot <slug> and let it run to completion. No iteration loop, no session state, no checkpoints — those concepts were removed.

Execution

1. Resolve the slug

INPUT = trim($ARGUMENTS)

IF INPUT is empty:
    pick = invoke Skill({ skill: "ideate", args: "--next" })
ELSE IF INPUT matches ^[A-Z]+-\d+$ OR ^[a-z][a-z0-9-]*$:
    SLUG = INPUT
    GOTO step 2
ELSE:
    pick = invoke Skill({ skill: "ideate", args: "--next --focus \"$INPUT\"" })

# Parse the ITEM line from ideator's --next output:
#   ITEM: <ID or title>
#   SCHEMA: <feature|bugfix|chore>
#   REASON: ...
#   PERSISTED: no
SLUG = first line matching `^ITEM:\s*(\S+)` from `pick`

IF SLUG is empty or unparseable:
    HALT — print: "Autopilot: ideator returned no actionable item. Run /ideate manually to seed the backlog."

2. Run the autopilot workflow to completion

orchestrator autopilot $SLUG

The autopilot subcommand resolves config/workflows/autopilot.yaml directly (each workflow file is a CLI subcommand: orchestrator autopilot <slug> == orchestrator run <slug> --schema autopilot). It seeds schema=autopilot, runs the produce → implement → complete spine straight through with no review gates, and drives the loop to completion via the in-process dispatch loop (orchestrator_next/run_loop.py).

Autopilot ends at the boundary — it does NOT merge. The loop archives the change on the feature worktree (archive-completed-change) and exits; the worktree and branch are left intact. Merging is a deliberate, separate action — orchestrator complete <slug> (/complete-feature or /approve-qa) — which merges unconditionally. Do not pause for confirmation during the autopilot run itself.

What This Skill Does NOT Do

• No iteration loop. One run, one feature.
• No session files, checkpoints, or rollups.
• Does not duplicate workflow logic — the orchestrator autopilot subcommand handles seeding, the dispatch loop, and complete.
• Does not pick the ticket itself — defers to ideator (or accepts an explicit slug).

Failure modes

• No actionable item — ideator returns nothing parseable: halt with the message above. User runs /ideate to seed backlog.
• Workflow fails mid-run — surfaces inside orchestrator autopilot; this skill exits with whatever the subcommand returns. No retry, no fallback.