Crate zeph_subagent 

Subagent management: spawning, grants, transcripts, and lifecycle hooks.

zeph-subagent provides the full lifecycle of sub-agent tasks within the Zeph agent framework. It covers:

• Definitions (SubAgentDef) — parse YAML/TOML frontmatter from .md files, validate names and permissions, and load from priority-ordered directories.
• Manager (SubAgentManager) — spawn, cancel, collect, and resume sub-agent tasks against a configurable concurrency limit.
• Grants (PermissionGrants) — zero-trust TTL-bounded permission tracking for vault secrets and runtime tool grants.
• Hooks (fire_hooks) — run shell commands at PreToolUse, PostToolUse, SubagentStart, and SubagentStop lifecycle events.
• Filter (FilteredToolExecutor) — enforce per-agent ToolPolicy and denylist on every tool call before it reaches the real executor.
• Transcripts (TranscriptWriter, TranscriptReader) — persist conversation history to JSONL files for session resume and auditing.
• Memory (ensure_memory_dir, load_memory_content) — resolve and inject persistent MEMORY.md content into the sub-agent system prompt.
• Commands (AgentCommand, AgentsCommand) — typed parsers for /agent and /agents slash commands.

Quick start

use std::sync::Arc;
use zeph_subagent::{SubAgentDef, SubAgentManager};

// Parse a definition from markdown frontmatter.
let content = "---\nname: helper\ndescription: A helpful sub-agent\n---\nYou are a helper.\n";
let def = SubAgentDef::parse(content).expect("valid definition");
assert_eq!(def.name, "helper");

// Create a manager with a concurrency limit of 4.
let _manager = SubAgentManager::new(4);

Re-exports

pub use command::AgentCommand;

pub use command::AgentsCommand;

pub use def::SubAgentDef;

pub use def::SubAgentPermissions;

pub use def::is_valid_agent_name;

pub use error::SubAgentError;

pub use filter::FilteredToolExecutor;

pub use filter::PlanModeExecutor;

pub use filter::filter_skills;

pub use grants::Grant;

pub use grants::GrantKind;

pub use grants::PermissionGrants;

pub use grants::SecretRequest;

pub use hooks::HookError;

pub use hooks::fire_hooks;

pub use hooks::matching_hooks;

pub use manager::SpawnContext;

pub use manager::SubAgentHandle;

pub use manager::SubAgentManager;

pub use manager::SubAgentStatus;

pub use memory::ensure_memory_dir;

pub use memory::load_memory_content;

pub use resolve::resolve_agent_paths;

pub use state::SubAgentState;

pub use transcript::TranscriptMeta;

pub use transcript::TranscriptReader;

pub use transcript::TranscriptWriter;

pub use transcript::sweep_old_transcripts;

Modules

command

Typed parsers for /agent and /agents slash commands.

def

Sub-agent definition parsing and loading.

error

filter

Tool and skill filtering for sub-agents.

grants

Zero-trust TTL-bounded permission grants for sub-agents.

hooks

Lifecycle hooks for sub-agents.

manager

Sub-agent lifecycle management: spawn, cancel, collect, and resume.

memory

Persistent per-agent memory backed by MEMORY.md files on the local filesystem.

resolve

Priority-ordered resolution of sub-agent definition directories.

state

transcript

JSONL-based transcript persistence for sub-agent conversations.

Structs

HookDef

A single hook definition.

HookMatcher

Tool-name matcher with associated hooks.

SkillFilter

Skill allow/deny filter for sub-agent definitions.

SubagentHooks

Per-agent frontmatter hook collections (PreToolUse / PostToolUse).

Enums

HookType

The type of hook — currently only shell command hooks are supported.

MemoryScope

Persistence scope for sub-agent memory files.

ModelSpec

Specifies which LLM provider a sub-agent should use.

PermissionMode

Controls tool execution and prompt interactivity for a sub-agent.

ToolPolicy

Tool access policy for a sub-agent.