Expand description
Run configuration: the agent roster, the shape of the graph, and the blindness / verification policy.
Discovery order (first hit wins):
--config <path><repo>/magi.toml<repo>/.magi/config.toml<config_dir>/magi/config.toml- built-in defaults, with the agent roster derived from the agent CLIs actually installed on this machine
Structs§
- Agent
Spec - One addressable agent in the roster.
- Blind
- Blindness policy.
- Config
- Top-level configuration.
- Daemon
- How the daemon loop itself behaves, as opposed to what one run does.
- Disk
- Disk hygiene: how hard magi is allowed to press on the machine’s free space.
- Graph
- Graph shape and limits.
- Merge
- Merge policy.
- Notify
- How the operator is told that a run is waiting on them.
- Prompts
- Project-specific text appended to each node’s prompt.
- Repos
- Where
magi reposandGET /api/reposlook for local checkouts. - Resolved
Roles - Roles resolved to concrete agent specs for one run.
- Roles
- Explicit role assignment. Empty lists are filled in by
Config::resolve_rolesby rotating the roster. - Talk
- Policy for
crate::talk, the standing conversation. - Update
- Self-update policy.
- Verify
- Shell commands that gate the winner.
Enums§
- Agent
Kind - Which CLI drives an agent.
- Delivery
- How the prompt reaches the agent process.
- Incomplete
Review Policy - What a review round does when a reviewer seat never answered.
- Leak
Policy - What to do when vendor-identifying text is found in material shown to judges.
- Merge
Mode - What to do with the winning branch.
- Merge
Style - How the winning branch is attached to the base branch: what
mode = "local"runs, and whatmode = "none"’s printed guidance tells the operator to run by hand. - Update
Mode - How magi keeps itself current.
Functions§
- which
- Is
programonPATH?