Struct AgentDefinition 

pub struct AgentDefinition {

    pub name: String,
    pub layer: AgentLayer,
    pub cli_tool: String,
    pub task: String,
    pub schedule: Option<String>,
    pub model: Option<String>,
    pub default_tier: Option<String>,
    pub capabilities: Vec<String>,
    pub max_memory_bytes: Option<u64>,
    pub budget_monthly_cents: Option<u64>,
    pub provider: Option<String>,
    pub persona: Option<String>,
    pub terraphim_role: Option<String>,
    pub skill_chain: Vec<String>,
    pub sfia_skills: Vec<SfiaSkillRef>,
    pub fallback_provider: Option<String>,
    pub fallback_model: Option<String>,
    pub grace_period_secs: Option<u64>,
    pub max_cpu_seconds: Option<u64>,
    pub pre_check: Option<PreCheckStrategy>,
    pub gitea_issue: Option<u64>,
    pub event_only: bool,
    pub project: Option<String>,
    pub evolution_enabled: bool,
    pub rlm_enabled: Option<bool>,
    pub bypass_kg_routing: bool,
    pub enabled: bool,
}

Definition of a single agent in the fleet.

Fields

name: String

Unique name (e.g., “security-sentinel”).

layer: AgentLayer

Which layer: Safety, Core, Growth.

cli_tool: String

CLI tool to use (maps to Provider).

task: String

Task/prompt to give the agent on spawn.

schedule: Option<String>

Cron schedule (None = always running for Safety, or on-demand for Growth).

model: Option<String>

Model to use with the CLI tool (e.g., “o3” for codex, “sonnet” for claude).

default_tier: Option<String>

Default KG tier concept for this agent (e.g., “review_tier” or “review”). When KG routing matches a different tier with confidence below the escalation threshold, the router falls back to this tier instead.

capabilities: Vec<String>

Capabilities this agent provides.

max_memory_bytes: Option<u64>

Maximum memory in bytes (optional resource limit).

budget_monthly_cents: Option<u64>

Monthly USD budget in cents (e.g., 5000 = $50.00). None means unlimited (subscription model).

provider: Option<String>

LLM provider for this agent (e.g., “openai”, “anthropic”, “openrouter”).

persona: Option<String>

Persona name for this agent (e.g., “Security Analyst”, “Code Reviewer”).

terraphim_role: Option<String>

Terraphim role identifier (e.g., “Terraphim Engineer”, “Terraphim Designer”).

skill_chain: Vec<String>

Chain of skills to invoke for this agent.

sfia_skills: Vec<SfiaSkillRef>

SFIA skills with proficiency levels.

fallback_provider: Option<String>

Fallback LLM provider if primary fails.

fallback_model: Option<String>

Fallback model if primary fails.

grace_period_secs: Option<u64>

Grace period in seconds before killing an unresponsive agent.

max_cpu_seconds: Option<u64>

Maximum CPU seconds allowed per agent execution.

pre_check: Option<PreCheckStrategy>

Optional pre-check strategy to gate agent spawns. If None, the agent always spawns (equivalent to Always).

gitea_issue: Option<u64>

Gitea issue number to post output to (optional).

event_only: bool

True if this agent must NOT be dispatched from comment mentions. Used for event-only agents like build-runner that are spawned only by handle_push with ADF_PUSH_* context. When true:

• handle_webhook_dispatch::SpawnAgent rejects any mention of this agent and logs an explicit rejection.
• The post-exit OutputPoster hook skips publishing stdout/stderr to gitea_issue even if it is somehow set, because event-only agents post their verdict via a different channel (e.g. the Gitea Commit Status API).

Default false keeps existing LLM agents mention-dispatchable.

project: Option<String>

Project this agent belongs to. Must match a Project.id when any projects are defined. None means the agent is global / legacy single-project mode; mixing per-project and global agents is rejected at load time.

evolution_enabled: bool

Enable evolution tracking for this agent (requires evolution feature and top-level evolution.enabled = true). Default: false.

rlm_enabled: Option<bool>

Enable RLM sandboxed code execution for this agent. When Some(true), the orchestrator injects RLM session info into the agent prompt, enabling the agent to request isolated code execution via terraphim_rlm. Default: None (disabled).

bypass_kg_routing: bool

If true, spawn_agent honours the explicit cli_tool and model on this definition and skips the KG tier-routing override block. Set by the quota-exit and wall-clock-timeout fallback respawns so the operator-chosen fallback provider is not overridden by the same tier-routing rule that selected the now-blocked primary.

Default false preserves existing behaviour for normal (non-fallback) spawns – KG tier routing continues to dominate static config.

enabled: bool

Whether this agent is enabled. Set to false by the config-error circuit-breaker after 3 consecutive ConfigError exits. Defaults to true.

Trait Implementations

impl Clone for AgentDefinition

fn clone(&self) -> AgentDefinition

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for AgentDefinition

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for AgentDefinition

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for AgentDefinition

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

where __S: Serializer,

Serialize this value into the given Serde serializer.