vigiles API reference
    Preparing search index...

    Interface AgentRunArgs

    The per-trial arguments handed to an AgentRunner.

    interface AgentRunArgs {
        cwd: string;
        effort?: string | number;
        env?: Record<string, string>;
        hasSettings: boolean;
        model: string;
        pluginDir: string | undefined;
        replaceEnv?: boolean;
        task: string;
        timeoutMs: number;
        tools: readonly string[];
    }
    Index

    Properties

    cwd effort? env? hasSettings model pluginDir replaceEnv? task timeoutMs tools

    Properties

    cwd: string
    effort?: string | number

    Reasoning-budget level for the run (claude --effort). Part of the MEASUREMENT, not a run knob: it changes the model's output distribution, not the sample size — so it lives on the spec next to model (never an env), and it is hashed into both the cache key and the eval lock. Deliberately string | number rather than a literal union: the binary accepts an alias map, is case-insensitive, and takes an integer budget, and its own valid set MOVED between builds (2.1.42 had no xhigh, 2.1.257 does) — a hard-coded union would reject a valid level after any upstream addition. A wrong value is caught at RUNTIME instead, by effortRejection, which is what the binary actually tells us. Omit for the harness default.

    env?: Record<string, string>

    Extra env layered over process.env for this run (e.g. VIGILES_INTERCEPT_TOOLS).

    hasSettings: boolean
    model: string
    pluginDir: string | undefined
    replaceEnv?: boolean

    When true, env is the COMPLETE spawn environment (an ephemeral run env from ephemeralRunEnv) — the runner does NOT prepend process.env, so the real $HOME / secrets are scrubbed. Default false: env is an overlay over process.env (the byte-identical-to-today path). Set only by an ephemeral env.

    task: string
    timeoutMs: number
    tools: readonly string[]

    Settings

    Member Visibility

    On This Page

    Properties