Struct Agent 

pub struct Agent {

    pub runtime: String,
    pub model: Option<String>,
    pub role_prompt: Option<RolePrompt>,
    pub permission_mode: Option<String>,
    pub autonomy: String,
    pub can_dm: Vec<String>,
    pub can_broadcast: Vec<String>,
    pub reports_to: Option<String>,
    pub on_rate_limit: Option<Vec<String>>,
    pub effort: Option<EffortLevel>,
    pub ultracode: bool,
    pub interfaces: Option<AgentInterfaces>,
    pub display_name: Option<String>,
    pub hooks: Vec<HookSpec>,
    pub mcps: BTreeMap<String, McpServer>,
    pub subagents: Vec<PathBuf>,
    pub skills: Vec<PathBuf>,
}

Fields

runtime: String

model: Option<String>

role_prompt: Option<RolePrompt>

permission_mode: Option<String>

autonomy: String

can_dm: Vec<String>

can_broadcast: Vec<String>

reports_to: Option<String>

on_rate_limit: Option<Vec<String>>

Override the global rate-limit hook chain for this agent.

effort: Option<EffortLevel>

Per-agent reasoning effort. Renders as EFFORT=<value> in the agent env file; the wrapper passes it to the runtime (e.g. claude --effort <value>). Strict enum: typos like hgih fail compose validation rather than silently falling back to the wrapper default.

ultracode: bool

#461: per-agent ultracode opt-in. When true, teamctl emits "ultracode": true into the agent’s Claude Code settings JSON (the file the wrapper passes via --settings); the resulting session behavior is Claude Code’s, not teamctl’s. ultracode is a Claude Code setting — not an effort level and not a CLI flag — so it’s orthogonal to effort: above (which rides the --effort flag) and the two compose freely. Like the enableAllProjectMcpServers key in render_claude_settings, the name is vendor-owned: verified against Claude Code 2.1.175, and a future rename would silently no-op. claude-only v1: on non-claude runtimes the settings file is skipped and a declared opt-in warns (codex/gemini have no equivalent). Defaults to false → the settings JSON is byte-unchanged for agents that don’t opt in.

interfaces: Option<AgentInterfaces>

Per-manager human-facing interfaces. Today’s only adapter is telegram; the shape is reserved for future adapters (discord, imessage, …) so a manager can declare every channel it speaks on in one place. Workers leave this unset.

display_name: Option<String>

T-160: optional human-friendly label rendered by the TUI in place of the agent id (roster, details header, mailbox row attribution, statusline). Absent → render the agent id (current behavior). Validation: non-empty, ≤64 chars, UTF-8 anything. The agent id stays canonical for routing, tmux session names, CLI args, and YAML cross-refs (can_dm, can_broadcast, reports_to) — display_name is render-time only.

hooks: Vec<HookSpec>

#383 Phase 2: per-agent Claude Code hooks, merged additively into the per-agent settings.json that render already builds, on top of the built-in interactive-prompt deny hook (which keeps precedence — see render::render_claude_settings). Commands are compose-root-relative paths, resolved like role_prompt. claude-only v1: declared on a non-claude-code agent they render nothing and render logs an “unsupported runtime” warning. Empty (the default) → settings unchanged.

mcps: BTreeMap<String, McpServer>

#383 Phase 4: per-agent MCP servers, merged into the rendered per-agent MCP config alongside the built-in team mailbox server (which stays unconditional and non-clobberable — a declared server named team is rejected at validate and skipped at render). Unlike hooks, MCP is runtime-agnostic: declared servers render for every runtime whose descriptor sets supports_mcp, and are skipped with a warning otherwise. Empty (the default) → MCP config unchanged.

subagents: Vec<PathBuf>

#383 Phase 3a: per-agent Claude Code sub-agents declared in compose. Each entry is a compose-root-relative path to a standard sub-agent markdown file (frontmatter name / description / optional tools / model, body → the sub-agent’s system prompt), resolved like role_prompt. render transforms the list into Claude Code’s --agents inline JSON so each agent gets its own sub-agents additively, on top of the project .claude/agents/ and the built-in sub-agents (verified: --agents adds, never replaces), without an arbitrary-path flag (the only cwd-stationary mechanism — see the Phase-1 spike). claude-only v1: declared on a non-claude-code agent they render nothing and render logs an “unsupported runtime” warning. Empty (the default) → no --agents flag.

skills: Vec<PathBuf>

#383 Phase 3b: per-agent Claude Code skills declared in compose. Each entry is a compose-root-relative path to a skill directory (the folder holding SKILL.md), resolved like role_prompt. render materializes a per-agent scope dir under state/agent-scope/<project>-<agent>/.claude/skills/ holding a symlink to each declared skill, and the wrapper passes that scope dir via claude --add-dir so the agent discovers its skills additively, on top of the project .claude/skills/ (verified: --add-dir adds, never replaces — see the Phase-1 spike §9-E1). claude-only v1: declared on a non-claude-code agent they materialize nothing and render logs an “unsupported runtime” warning. Empty (the default) → no --add-dir flag.

Implementations

impl Agent

pub fn telegram(&self) -> Option<&TelegramConfig>

Convenience: pull the manager’s Telegram config out of interfaces.telegram without forcing every callsite to handle the nested options.

Trait Implementations

impl Clone for Agent

fn clone(&self) -> Agent

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Agent

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for Agent

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for Agent

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

where __S: Serializer,

Serialize this value into the given Serde serializer.