Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add haiggoh/run-to-completion --skill autopilotgit clone --depth 1 https://github.com/haiggoh/run-to-completionWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/haiggoh/run-to-completion/autopilot)<a href="https://agentmods.dev/skills/haiggoh/run-to-completion/autopilot"><img src="https://agentmods.dev/badge/skills/haiggoh/run-to-completion/autopilot.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00097 | $0.00888 |
| Opus 5 | $0.00048 | $0.00444 |
| Sonnet 5 | $0.00019 | $0.00178 |
| Haiku 4.5 | $0.00010 | $0.00089 |
Grade A, and why
autopilot scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 50 lines — stays where its author put it; the contents beside it link to each section on GitHub.
autopilot — the entry point for a whole unattended run
This skill orchestrates the complete run. It sequences three phase skills and adds the up-front kickoff. It does not duplicate their content.
Kickoff
Do this once before triage:
-
Establish the resource picture. Find out what you actually have to spend — time, a request or token quota, a cost cap if one exists at all — by checking, not by assuming. Many setups have no cap, and in those the answer is simply "no limit to plan around". Never conclude you are blocked without looking: if the user's prompts are being answered, the pipe is working.
-
Warm any delegate. Start any local-execution capability in the background so the first delegation is not a cold start. Capture whatever address or port the warm-up reports.
-
Ask blocking questions about THE RUN — and only about the run. Scope, and permission for any push, publish, or confirmation-gated action. Ask for the ship-loop depth as one question — cleared through push, through tag, or through release — rather than a bare yes to “may you push”, because the loop's later steps publish and a run with nobody present cannot widen its own permission when it reaches one. Do not stop for those again.
The asymmetry that makes this work: kickoff questions are about the run; a question about a specific item discovered during triage becomes a gate on that item, never a kickoff question. Otherwise kickoff degenerates into interrogating sixty items one by one, which defeats the entire point of "clear what you can while I am away" — the user walks away precisely so they do not have to answer things.
The corollary cuts the other way and is just as important: a known question is a gate, and work sitting before that question is actionable. So an item that ends in a question is not thereby excluded from the run.
Contrast this with an attended run-to-completion, where the user is present: there, prefer surfacing and clearing gates as early as you can, because an answer is cheap and immediately unblocks work. Same plugin, opposite instinct, decided by whether anyone is there to answer.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday Changed 74a1377db873
- 4d ago Changed · +6 lines dcfe1b016530
- 8d ago First seen · 44 lines · 97 tokens per session scan A 76cafbfe3c69
autopilot is a skill published in the GitHub repository haiggoh/run-to-completion (1 stars, last pushed yesterday), licensed MIT. It adds 97 tokens to every session and 888 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
tasks
A task-breakdown tool that turns a plan, specification, feature brief, bug report, technical-debt item, or chore into small TASK-NNN work items. Technical debt means postponed cleanup or design work in a codebase.
continue
Autonomous work — find and execute ready tasks.
prd
A Product Requirements Document (PRD) writer for large software initiatives. A PRD records the problem, goals, requirements, assumptions, and evidence before implementation begins.
roadmap
An implementation plan made from a technical specification, organized into phases, milestones, and dependencies.
debt
A command for showing and closing unanswered questions attached to project documents such as requirements, specifications, features, and designs. It can show questions for the whole project or for one document.
pr
A command for managing pull requests on GitHub or Bitbucket Server. A pull request is a request for other developers to review and merge changes into a project branch.