@agnishc/edb-subagents

Claude Code-style autonomous sub-agents for pi. Spawn specialized agents that run in isolated sessions — each with its own tools, system prompt, model, and thinking level. Run them in foreground or background, steer them mid-run, resume completed sessions, and define your own custom agent types.

Forked from tintinweb/pi-subagents and extended as part of the edb monorepo.

Install

pi install npm:@agnishc/edb-subagents

Subagents V2 opt-in

V2 is available behind an explicit engine setting while V1 remains the default. Add this to .pi/subagents.json:

{
  "engine": "v2"
}

V2 uses persistent Pi sessions, recursive foreground/background agents, logical wait/resume, nested questions, bounded concurrency, atomic recovery state, and an optional versioned RPC connection to @agnishc/edb-todo. Child sessions load exact tool allowlists and no parent extensions. Project agent definitions are loaded only for trusted projects.

V2 management commands are /agents tree, /agents questions, /agents show <agentId>, /agents stop <agentId>, /agents dispose <agentId>, /agents recovery, /agents outbox, and /agents diagnostics.

The V1-only worktree, scheduler, memory, group-join, and bridge behaviors listed below are not part of V2.

Features

The following features describe the default V1 engine:

• Live widget — persistent above-editor widget with animated spinners, live tool activity, token counts, and context utilisation
• Parallel background agents — spawn multiple agents concurrently with automatic queuing
• Conversation viewer — /agents → select a running agent to see its full conversation live
• Custom agent types — define agents in .pi/agents/<name>.md with YAML frontmatter
• Mid-run steering — inject messages into running agents via steer_subagent
• Session resume — continue a previous agent's full conversation context
• Graceful turn limits — agents wrap up cleanly before hard abort
• Git worktree isolation — run agents in isolated repo copies
• Persistent agent memory — three scopes: project, local, user
• Scheduled agents — cron, interval, and one-shot scheduling
• Cross-extension RPC — spawn/stop agents from other extensions via pi.events

Quick start

Agent({
  subagent_type: "Explore",
  prompt: "Find all files that handle authentication",
  description: "Find auth files",
  run_in_background: true,
})

Custom agents

Create .pi/agents/<name>.md:

---
description: Security Code Reviewer
tools: read, grep, find, bash
model: anthropic/claude-opus-4-6
thinking: high
max_turns: 30
---

You are a security auditor. Review code for vulnerabilities...

Commands

Command	Description
/agents	Interactive agent management — view running agents, manage types, create new agents, settings

Attribution

Core implementation by tintinweb, MIT licensed.