docs
the harness, documented.
These are the harness's own documents, mirrored from the repository that builds it and rendered here as static pages. Nothing on them is written for the website: if a claim is wrong, it is wrong in Colonizer-dev/harness first, and this mirror follows.
- Install What your machine needs, how to build and run the harness, the first run, where things live and how to update. install.md
- Updates Which version you run, how the update check works, and how to move to a newer one without losing colonies. updates.md
- Cockpit The web UI, view by view: signing in, what each page shows, how to reach it and where it stops. cockpit.md
- A colony's life One colony from launch to cleanup: what each feature does, how to switch it on or off, its default, and where it stops. colonies.md
- CLI The colonizer command: starting the mothership, and driving colonies from a terminal, here or across a tailnet. cli.md
- MCP server colonizer mcp: the harness as tools for an MCP client, scoped by the token it runs with. mcp.md
- Loops A saved prompt on a repository that launches a colony on a schedule. loops.md
- Vision What Colonizer is for, the vocabulary the rest of the docs use, and the principles the design answers to. vision.md
- Decisions What has been decided against, and why. decisions.md
- Gaps What the design and this site depict that the code does not do yet. gaps.md
- Configuration Where settings live, the master key and the system keychain, the limits on each colony, the spend journal and colonizer.toml. configuration.md
- Architecture The moving parts on your machine and inside a colony, the module table, the session lifecycle and the mesh. architecture.md
- Protocol The wire contracts: runner stdio, the agentd API inside the VM, the harness API the browser talks to. protocol.md
- Model providers Which provider connections work with which agent module, and the switches that take tools away from a colony. providers.md
- Skill packs The layout of a skillset directory, how one is validated, and the downloadable ones. skill-packs.md
- Runner authoring The checklist a new agent module has to pass before it ships. runner-authoring.md
- Session storage Where colony records and evidence live, and the contract a different store has to keep. session-store.md
- Remote access The opt-in my.colonizer.dev link to your own cockpit: the tunnel contract between the mothership and the relay. remote-tunnel.md
- Remote access review The security review of the relay and the tunnel before the relay is deployed. remote-access-review.md
- Outposts The seam between deciding and running colonies, the local side of it, and what remote outposts still need. outposts.md
- Fleets Motherships that pair with each other: the invite and confirmation code, the shared fleet view, and the history a member can sync to its owner. fleet.md
- Hosted The contract a hosted Colonizer would have to keep: the same mothership, run somewhere else. A design gate, not a feature. hosted.md
- Trust model Where each credential lives, and what a colony holds instead: placeholders, never the real token. trust-model.md
- Audit What an external audit confirmed, what it found instead, and what has to be true before unattended work. audit.md
- Escape vectors Each known way out of a colony, tried against the real sandbox, with its verdict and the mechanism that blocks it. escape-vectors.md
- Bench The fixed tasks colonies are scored against, and how to run the benchmark. bench.md
- Trajectory monitor How a colony's event log is audited for shortcuts, splitting resolved from clean resolved. trajectory-monitor.md
- Red team Hunter swarms that raid one repository while the nest is empty, and the merged report they leave behind. red-team.md
- Security hunters The Strix and Shannon hunter modules: what installs today, and why no scan runs yet. security-hunters.md
- Burn-down Spending a weekly token plan to a reserve with paced bug-hunt colonies. burn-down.md
- Sandbox network What a colony's network policy allows and denies, as the harness drives microsandbox. sandbox-network.md
- Path policy Worktree paths a colony may not read, and paths it may read but never rewrite. path-policy.md
- Boundary vs guidance Which controls the kernel or the host enforces, and which are only guidance to the agent. boundaries.md
- Prompt screening The publish-time check for hidden characters in what a colony is about to push. prompt-screening.md
- Observability Logs, metrics and one trace per colony, sent over OTLP to a backend you name. Off until you configure it. observability.md
- Jev The optional external second opinion at named decision points: its modes, what it is sent, and how to keep it off. jev.md
- Changelog What changed in each release, newest first, with the pull request behind each line and anything that could cost work called out. changelog.md
- Live map What a mothership sends when the live map is switched on, what the service keeps and for how long, and how to keep it off. telemetry.md
- Usage data The anonymous batch of counts the harness builds about its own use, what it contains, and how to switch it off. usage-data.md
The mirror
| repository | Colonizer-dev/harness |
| ref | v0.2.14 |
| commit | f59bdf7 |
| taken | 2026-10-08T06:12:25Z |
Diagrams are rendered to SVG at build time, so these pages carry no script
and fetch nothing. Re-sync with tools/sync-docs.sh, then
node tools/docs.js.