Hyper autopilot

Hyper-autopilot — autonomous-loop runner (scheduled or immediate)

Thin driver over the hyperclaude plugin. Assemble a per-round loop prompt — /hyperclaude:hyper-auto → ff-merge to main → push → pick next task → repeat — and fire it now or on a schedule. hyper-auto owns the plan→implement chaining and its between-phase safety stop; autopilot only adds scheduling + merge/push/next-task orchestration on top.

• Prerequisite: if hyperclaude is not installed there is nothing to drive — STOP and say so. Never hand-roll the plan/implement steps.
• Single round only (no auto-pickup of the next task): skip this skill, run /hyperclaude:hyper-auto directly. Autopilot's value is the multi-round loop.
• Remote/cloud execution: use /schedule (RemoteTrigger), not this.

Modes

• Immediate — "now" / "right away" / no time given: start executing the assembled prompt inline this session.
• Scheduled — "in N hours" / "tomorrow 8am" / "at 03:00": fire once via CronCreate. The session must stay alive and idle until then (closing the terminal kills it; locking the laptop is fine).

Ambiguous phrasing → ask which, don't guess.

Procedure

1. Inputs. Round-1 task (one paragraph — hyper-auto expands it). Optional: follow-up candidates, extra constraints.

2. (Scheduled only) Resolve time + build cron. Read live time (date -u +%Y-%m-%dT%H:%M:%SZ, date +%Y-%m-%d) — never a stale conversation anchor. Parse the user's "when" against it and echo the resolved local time back ("fires: 2026-05-21 04:00 KST") to catch a misparse. Build a one-shot local-tz 5-field cron M H DoM Mon DoW, recurring: false, off-minute (:07/:13/:22) unless an exact time was given.

3. Precheck push. git remote -v, gh auth status. If the remote is unset or auth is broken, surface it before dispatching. If main is protected / PR-required so a direct push is blocked, surface that too — default delivery is ff-merge + push origin main; switch to PR-based only if the user confirms.

4. Assemble + confirm. Fill the template below: {{round1}} verbatim; {{candidates}} = the user's list or, if none, the backlog fallback noted under the template; include {{extra}} only if the user gave extra constraints. Show the full assembled prompt (scheduled: also the firing time + cron) and get an explicit "go" — this commits and pushes, so never auto-fire.

5. Dispatch.

   • Scheduled → CronCreate({ cron, recurring: false, durable: false, prompt }). Echo the job id (for CronDelete), the local firing time, and the "session must stay alive & idle" caveat.
   • Immediate → do NOT call CronCreate. Treat the assembled prompt as the active directive and start Round 1 now by invoking /hyperclaude:hyper-auto. Keep the assembled rules in scope verbatim — do not paraphrase them away.

Loop prompt template (substitute in step 4)

[Auto trigger — start autonomous workflow]

## Round 1 task
{{round1}}

## Per-round 3 steps
1. **/hyperclaude:hyper-auto** — chain `hyper-plan-loop` → `hyper-implement-loop` for this round's task. It stops between phases on plan-loop cap-reached or terminal revise-regression; treat that as a round failure (see Stop conditions).
2. **Fast-forward merge to main** — if on a work branch, checkout main then `git merge --ff-only`.
3. **git push origin main** — no force push, no `--no-verify`; on hook failure, fix the root cause and create a new commit.

After step 3 → autonomously pick the next task and start the next round.

## Next-task candidates (priority order)
{{candidates}}

## Stop conditions
- Context limit near → finish the current round and stop.
- Any step needs a destructive git op (force push / hard reset / branch -D) → stop and report.
- Push fails (remote rejection, auth) → stop and report.
- `hyper-auto` ends non-clean for either inner loop (plan-loop cap-reached / terminal revise-regression, or implement-loop with open Blocker/Major) → stop. Don't start the next round; don't retry the same task.

## Rules
- Respect the project's CLAUDE.md / AGENTS.md invariants, comment policy, and code style.
- Honor the project's existing quality gates (tests, lint/typecheck, coverage thresholds from its CLAUDE.md / CI). Never lower or skip a gate to make a round pass.
- End each round: append a one-line summary to `.hyperclaude/autonomous-log.md` (create if missing).
- Start each round: read the most recent `.hyperclaude/` artifacts (plans/, code-reviews/, plan-reviews/) for context.
- Skip visual verification (Playwright/manual) in autonomous mode — if the plan calls for it, note as unverified risk and proceed.

Substitution notes. {{candidates}} fallback when the user gives none: "After Round 1 pick the next task autonomously from the project's own backlog — README roadmap / TODO, open items in CLAUDE.md/AGENTS.md — and the most recent .hyperclaude/plans/ artifacts. If nothing is obviously next, end and report." If {{extra}} is empty, drop that section entirely (no trailing blank header).