Guides
The permission model, the config surface, running ari headless, and the working surfaces: edits and diagnostics, skills, hooks, project memory, and MCP.
Task-focused pages: how the permission pipeline decides what runs and the safety floor no mode can cross, how config layers resolve from defaults to flags, and how to drive ari from a script or a CI job with ari -p and the exit-code contract. Then the working surfaces: edits shown as diffs with language-server diagnostics folded in, skills and slash commands, hooks and workspace trust, project memory in ARI.md, and the MCP bridge.
The permission model
How the ordered pipeline decides what runs, the four modes, the session grants, and the safety floor no mode can cross.
Headless with ari -p
Run one turn and exit, stream the event schema with --json, take a prompt from stdin, and branch on the exit code in CI.
Edits, diffs, and diagnostics
How ari shows a change as a unified diff and folds language-server diagnostics into the edit result so it can catch and fix its own mistakes.
Skills and slash commands
Package a repeatable procedure as a skill the ant can invoke by name or you can run as a slash command, with the full body loaded only on use.
Hooks and workspace trust
Run your own commands at lifecycle points, and the trust gate that keeps a cloned repo's hooks from running until you say so.
Project memory with ARI.md
Give the ant standing house rules for a repo in an ARI.md file it reads every session, with a size cap so a rule you wrote is a rule in force.
The MCP bridge
Attach an MCP server's tools to ari through an mcp.toml, with schemas loaded on demand and every call gated and treated as untrusted content.