Struct RunLimits 

pub struct RunLimits {
    pub max_steps: u32,
    pub timeout_ms: u64,
    pub max_step_retries: u32,
    pub terminal_app: Option<String>,
    pub workflow_id_for_memory: Option<String>,
    pub memory_db_path: Option<String>,
}

Budget limits for a single agent run.

These are the only knobs the agent loop itself consumes. Callers that used to set enable_vision / enable_decomposition / self_heal / enable_notebook are expected to stop — those behaviors are now implicit in the canonical agent loop.

Fields

max_steps: u32

Total step budget across all sub-goals. A step that retries 3 times still counts as 1 step.

timeout_ms: u64

Wall-clock deadline in milliseconds. Measured from GoalRunner::run start, not per sub-goal.

max_step_retries: u32

Per-step retry cap. Default: 3. See the 3-strike rule in docs/canonical-agent-plan.md.

terminal_app: Option<String>

App name the goal should ultimately land its output in (e.g. "Numbers" for a spreadsheet scenario). When set, the runner enforces a phase-gate: once steps_used >= max_steps / 2 with no terminal-app work recorded (no write_cells/save_document action, and frontmost app != terminal_app), it injects a synthetic history record telling the planner its next batch MUST begin with activate_app(terminal_app). On a second ignore the runner auto-dispatches the activation itself.

None = no phase-gate enforcement (legacy behaviour).

workflow_id_for_memory: Option<String>

PR2 opt-in: when set together with memory_db_path, the canonical runner writes a final outcome memory under this workflow_id after the run terminates (success or failure). Defaults to None — no memory writes happen if the caller doesn’t ask for them.

Mirrors the cel_perceive start { enable_memory, workflow_id } opt-in for sessions; here it’s at the runner level so any caller of CanonicalGoalRunner::run (CLI, MCP, eval harness, worker) can opt in independently.

memory_db_path: Option<String>

PR2 opt-in: SQLite path the canonical runner uses to write the final outcome memory. Required alongside workflow_id_for_memory for the auto-write to fire.

Trait Implementations

impl Clone for RunLimits

fn clone(&self) -> RunLimits

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for RunLimits

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for RunLimits

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for RunLimits

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for RunLimits

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.