Colonizer v0.2.6

Colonizer v0.2.6

Colonies recover from the failures that have no work behind them

A colony that stopped over a provider blip, a status that fell out of step, a paused parent or an
expired sign-in used to sit in "Needs you" until a person noticed. In v0.2.6 the mothership handles
these itself:

• A transient provider error is retried with backoff, not held (#980). A turn that ends on a
  gateway 5xx, a 429 or 529, an unreachable or overloaded provider, a timeout or a dropped
  connection parks the colony (slot released, worktree kept) and continues it after 2, 5, 10 and
  20 minutes. Only when the attempts run out is it held, and then the message names the provider's
  own error instead of telling you to press Create PR. The Watchdog setting
  provider_retry_max_attempts (default 4, 0 to turn it off) sets the budget.
• A stale "waiting for an answer" is reconciled (#981). A colony left in
  waiting_for_answer with no question pending is set back to idle on the next watchdog tick,
  and an answer to one question no longer closes another that is still open, such as a subagent's
  exec-policy approval.
• Stacked colonies wait for their parent instead of failing in a cascade (#982). A colony
  started with after waits while its parent is stopped or parked, and moves onto the
  grandparent when the parent fails, rather than being retired with everything stacked behind it.
• Every Claude account is health-checked (#983). Settings → Connections shows whether each
  account is reachable, its token was rejected, or Anthropic could not be reached, and when it was
  last checked.
• An expired sign-in holds its colonies and tells you once (#984). A 401 or 403 from
  Anthropic marks the account and parks its colonies as waiting_for_account, without spending
  their retries. You get one notification per account, the cockpit shows a banner with a Sign in
  button, colonizer list prints a warning, and the colonies resume by themselves once you sign
  in again. Usage limits (429) keep the existing quota pause.

How each one works, and its limits, is in
docs/colonies.md, Recovering on its own.

What's new in v0.2.6

The rest, each backed by an entry in
CHANGELOG.md:

• Full autonomy (YOLO). A second autonomy setting: the judge with no max_answers cap,
  answering every question at or below its risk ceiling for as long as the colony runs. It still
  only picks among the agent's own options and never overrides a deny. Saving either autonomy
  setting without a model is now refused with a clear error (#776).
• An operator's Markdown vault, read-only in every colony. Point Colonizer at a local vault (an
  Obsidian vault, say), allowlist folders with a scope, and colonies read a filtered,
  secret-scrubbed snapshot at /colonizer/vault/ with an INDEX.md. Off by default (#777).
• Groundwork for OTLP telemetry export. A new colonizer-observability crate with one export
  policy every value goes through. Nothing sends telemetry yet, and the mothership does not link it.
  The secret redactor moved into its own crate, colonizer-redact, which is now published to
  crates.io next to the harness (#844).
• The protocol and loops references are split into one page per area and per loop, with the old
  protocol.md#… and loops.md#… links still landing (#826).
• Vendored plugins refreshed: ecc v2.2.3, caveman v3.1.0, google-skills 1d77046 and
  archify 3a5e785 (#712).

Install

curl -fsSL https://colonizer.dev/install.sh | sh

Prebuilt bundles are published for Linux x86_64 (with KVM) and Apple Silicon Macs
(darwin-arm64); there is no Intel-Mac or Windows build. What one machine needs:

• Linux: /dev/kvm readable and writable by your user (sudo usermod -aG kvm "$USER", then log
  out and back in) and a native Claude Code install, which colonies run.
• Either platform: git, gh, curl and tar (with xz); on Apple Silicon the installer also
  fetches the Linux build of Claude Code a colony needs.

Run colonizer afterwards; it prints a sign-in link and opens it. The full walkthrough — install
options, first run, where things live, how to update — is in
docs/install.md.

Security status

Colonizer is alpha and not ready for unattended work on sensitive repositories, real credentials
included. An external audit of v0.1.3 (commit d89ce76, 17 September 2026) found four ways a
colony could cross into the host; they are filed as draft advisories (F01–F04) and are not fixed
yet, and its four release checkpoints (G1–G4) are all uncleared. Controls added since — scoped
tokens, remote-access fencing, sensitivity-aware routing, gateway audit lines, log and secret
redaction, path policy, egress and exec policies — do not close those findings, so attended work is
what the harness is good for today. The live status is
docs/audit.md and its tracking
issue (#91); report vulnerabilities privately
through SECURITY.md.

Prebuilt Colonizer for Linux x86_64 (with KVM) and Apple Silicon Macs, built from 564c794.

curl -fsSL https://colonizer.dev/install.sh | sh

The installer checks the archive against SHA256SUMS. It fetches the Claude Agent SDK from the npm registry, and on a Mac the Linux build of Claude Code, from Anthropic's own channels: neither is in the release. Guide: https://colonizer.dev/docs/install

Each artifact, SHA256SUMS included, also carries a build-provenance attestation recording the commit that built it. Verify one with:

gh attestation verify colonizer-<platform>.tar.gz --repo Colonizer-dev/harness

Here is linux-x86_64 or darwin-arm64.

ee6bded8c17b2f5c26e9c0c69dd9957b5ffdd950d24c9e483e4348691490ac10  colonizer-darwin-arm64.tar.gz
1f7a3048ebfec8dc5a02c71462c7c974e0562d9f4f4476e10936c41cf2359b72  colonizer-linux-x86_64.tar.gz
12eeacb0c7248f4074b6343a1d787182a9ab18590837173dae19306ff33e85aa  install.sh