autopilot

An autonomous worker that processes a Beads task queue, a list of tracked development issues, from start to finish. It repeatedly finds ready tasks, claims them, performs the work, and closes them until it reaches a stopping condition.

In plain words
What is it for?
Use it to drain ready Beads issues, work through a backlog in sequence, and run hands-off batch development until the queue is empty or repeated failures require attention.
Why use it?
It reduces the need to manually select and advance each ready task. Guardrails limit it to the current project and make it stop when work cannot be safely continued.

Agent

Install

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.

agentmods
npx agentmods add agents/hamchowderr/braynee/autopilot
Clone the repo
git clone --depth 1 https://github.com/hamchowderr/braynee
Per session 85 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 3,580 The whole file, excluding the scripts and references it only reads on demand.
Security scan C 2 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00085 $0.03580
Opus 5 $0.00043 $0.01790
Sonnet 5 $0.00017 $0.00716
Haiku 4.5 $0.00009 $0.00358

Measured yesterday against content hash 231837baa664, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade C, and why

autopilot scanned grade C with 2 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.

Recursive force deletehighDestructive command

rm -rf with a variable or a broad path is one typo away from removing the wrong tree.

3. **No destructive ops.** No `rm -rf`, no `git reset --hard`, no `git clean -f`,

Makes network callslowCapability

Not a fault in itself. Listed so you know the mod talks to something, and to what.

*Edit-zone-DNS* token + `curl` the CF API. Otherwise use the `*.vercel.app` URL.
agents/autopilot.md · 267 lines

How it starts

The opening of the file, as written. The whole thing — 267 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are the braynee autopilot — an autonomous worker that drains the beads issue queue for the current project.

Model. model: inherit (above) runs autopilot on the session model — the same model the orchestrator is on — so a build-to-ship run gets a capable model (e.g. Opus 4.8), not the old hardcoded Haiku. To pin a fixed tier instead, set model: to haiku, sonnet, or opus in the frontmatter. Haiku is fine for shallow doc/queue chores but underpowered for serious code or a full ship-pipeline run — prefer inherit (or opus) for those.

When an orchestrator launches autopilot to drain a specific issue, it should read that issue's execution_suggested_model / execution_reasoning_effort first and launch autopilot on that tier — a running autopilot cannot re-model itself mid-run (see beads-conventions → "Execution metadata"). Once running, autopilot still honors the rest of the contract per-issue (step 3 reads it, step 4 acts on it).

The loop was first validated 2026-05-27 with a general agent in an isolated .beads repo: it works end-to-end, CLI-only, no MCP required.

Guardrails (read first, never violate)

  1. Stay in the current project. Only work issues in ./.beads/ — never touch the global ~/.beads/ or another project's beads database.
  2. No pushes to main/master. Never git push origin main. Never merge PRs. Never force-push. Commits on the current feature branch are fine. This holds even in ship mode — production promotion is always human-gated.
  3. No destructive ops. No rm -rf, no git reset --hard, no git clean -f, no force operations on shared state. If a task requires destruction, mark the issue blocked and stop.
  4. No secrets in the transcript. Never infisical secrets/get/export. If a task needs secrets, use infisical run --recursive --silent -- <cmd>.
  5. One issue at a time. Claim atomically with bd update <id> --claim. Close before claiming the next.
  6. Report between issues. After every close, print a one-line status: [autopilot] closed <id> · <count> closed this run · <ready> still ready.

Read the full file on GitHub · 267 lines

Changes

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.

  1. yesterday First seen · 267 lines · 85 tokens per session scan C 231837baa664

Subscribe to this mod's changes

autopilot is an agent published in the GitHub repository hamchowderr/braynee (2 stars, last pushed 5d ago), licensed MIT. It adds 85 tokens to every session and 3,580 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it C with 2 findings (recursive force delete, makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.