What if multiple agents in different terminals sharing a folder could talk to each other like they're in a chat room? Join, see who's online and what they're doing. Claim tasks, reserve files, send messages. An extension for Pi coding agent — install it and go. No daemon, no server, just files.
pi install npm:pi-messengerCrew agents ship with the extension (crew/agents/*.md) and are discovered automatically. The pi-messenger-crew skill is auto-loaded from the extension. Workers can load domain-specific crew skills on demand during task execution.
To show available crew agents:
npx pi-messenger --crew-installTo customize an agent for one project, copy it to .pi/messenger/crew/agents/ and edit it.
To remove the extension:
npx pi-messenger --removeTo remove stale crew agent copies from the shared legacy directory (~/.pi/agent/agents/):
npx pi-messenger --crew-uninstallOnce joined (manually or via autoRegister config), agents can coordinate:
pi_messenger({ action: "join" })
pi_messenger({ action: "reserve", paths: ["src/auth/"], reason: "Refactoring" })
pi_messenger({ action: "send", to: "GoldFalcon", message: "auth is done" })
pi_messenger({ action: "release" })
pi_messenger({ action: "leave" })For multi-agent task orchestration from a PRD:
pi_messenger({ action: "plan" }) // Planner analyzes codebase, creates tasks
pi_messenger({ action: "work", autonomous: true }) // Workers execute tasks in waves until done
pi_messenger({ action: "review", target: "task-1" }) // Reviewer checks implementationLiving Presence - Status indicators (active, idle, away, stuck), tool call counts, token usage, and auto-generated status messages like "on fire" or "debugging...". Your agent name appears in the status bar: msg: SwiftRaven (2 peers) ●3
Activity Feed - Unified timeline of edits, commits, test runs, messages, and task events. Query with { action: "feed" }.
Discovery - Agents register with memorable themed names (SwiftRaven, LunarDust, OakTree). See who's active, what they're working on, which model and git branch they're on.
Messaging - Send messages between agents. Recipients wake up immediately and see the message as a steering prompt.
File Reservations - Claim files or directories. Other agents get blocked with a clear message telling them who to coordinate with. Auto-releases on leave or exit.
Stuck Detection - Agents idle too long with an open task or reservation are flagged as stuck. Peers get a notification.
Human as Participant - Your interactive pi session appears in the agent list with (you). Same activity tracking, same status messages. Chat from the overlay.
/messenger opens an interactive overlay with agent presence, activity feed, and chat:
Chat input supports @Name msg for DMs and @all msg for broadcasts. Text without @ broadcasts from the Agents tab or DMs the selected agent tab.
| Key | Action |
|---|---|
Tab / ← → |
Switch tabs (Agents, Crew, agent DMs, All) |
↑ ↓ |
Scroll history / navigate crew tasks |
Enter |
Send message |
Esc |
Close |
Crew turns a PRD into a dependency graph of tasks, then executes them in parallel waves.
Crew logs are per project, under that project's working directory: .pi/messenger/crew/. For example, if you run Crew from /path/to/my-app, the planner log lives at /path/to/my-app/.pi/messenger/crew/planning-progress.md.
- Plan — Planner explores the codebase and PRD, drafts tasks with dependencies. A reviewer checks the plan; the planner refines until SHIP or
maxPassesis reached. History is stored inplanning-progress.md. - Work — Workers implement ready tasks (all dependencies met) in parallel waves. A single
workcall runs one wave.autonomous: trueruns waves back-to-back until everything is done or blocked. Each completed task gets an automatic reviewer pass — SHIP keeps it done, NEEDS_WORK resets it for retry with feedback, MAJOR_RETHINK blocks it. Controlled byreview.enabledandreview.maxIterations. - Review — Manual review of a specific task or the plan:
pi_messenger({ action: "review", target: "task-1" }). Returns SHIP, NEEDS_WORK, or MAJOR_RETHINK with detailed feedback.
No special PRD format required — the planner auto-discovers PRD.md, SPEC.md, DESIGN.md, etc. in your project root and docs/. Or skip the file entirely:
pi_messenger({ action: "plan", prompt: "Scan the codebase for bugs" })
// Plan + auto-start autonomous work when planning completes
pi_messenger({ action: "plan" }) // auto-starts workers (default)Tasks form a dependency graph. Independent tasks run concurrently:
Wave 1: task-1 (no deps) ─┐
task-3 (no deps) ─┤── run in parallel
│
Wave 2: task-2 (→ task-1) ─┤── task-1 done, task-2 unblocked
task-4 (→ task-3) ─┘── task-3 done, task-4 unblocked
Wave 3: task-5 (→ task-2, task-4) ── both deps done
The planner structures tasks to maximize parallelism. Foundation work has no dependencies and starts immediately. Features that don't touch each other get separate chains. Autonomous mode stops when all tasks are done or blocked.
Workers follow the same join/read/implement/commit/release protocol regardless of the task — what changes between tasks is domain knowledge. Crew skills let workers acquire that knowledge on demand.
Skills are discovered from three locations (later sources override earlier by name):
- User skills —
~/.pi/agent/skills/(pi's standarddir/SKILL.mdformat) - Extension skills —
crew/skills/within the extension (flat.mdfiles) - Project skills —
.pi/messenger/crew/skills/in your project root (flat.mdfiles)
The planner sees a compact index of all discovered skills and can tag tasks with relevant ones. Workers see tagged skills as "Recommended for this task" with the full catalog under "Also available", and load what they need via read(). Zero tokens spent until a worker actually needs the knowledge.
To add a project-level skill, drop a .md file in .pi/messenger/crew/skills/:
---
name: our-api-patterns
description: REST API conventions for this project — auth, pagination, error shapes.
---
# API Patterns
Always use Bearer token auth. Paginate with cursor-based `?after=` params.
Error responses use `{ error: { code, message, details? } }` shape.Any skills you already have in ~/.pi/agent/skills/ are automatically available to crew workers — no setup needed.
Crew spawns multiple LLM sessions in parallel — it can burn tokens fast. Start with a cheap worker model and scale up once you've seen the workflow. Add this to ~/.pi/agent/pi-messenger.json:
{ "crew": { "models": { "worker": "claude-haiku-4-5" } } }The planner and reviewer keep their frontmatter defaults; only workers (the bulk of the spend) get the cheap model. Override per-role as needed:
{
"crew": {
"models": {
"worker": "claude-haiku-4-5",
"planner": "claude-sonnet-4-6",
"reviewer": "claude-sonnet-4-6"
}
}
}Model strings accept provider/model format for explicit provider selection and :level suffix for inline thinking control. These work anywhere a model is specified — config, frontmatter, or per-task override:
{
"crew": {
"models": {
"worker": "anthropic/claude-haiku-4-5",
"planner": "openrouter/anthropic/claude-sonnet-4:high"
}
}
}The :level suffix and the thinking.<role> config are independent — if both are set, the suffix takes precedence and the --thinking flag is skipped to avoid double-application.
Full config reference (all fields optional — only set what you want to change):
{
"crew": {
"concurrency": { "workers": 2, "max": 10 },
"coordination": "chatty",
"models": { "worker": "claude-haiku-4-5" },
"review": { "enabled": true, "maxIterations": 3 },
"planning": { "maxPasses": 1 },
"work": {
"maxAttemptsPerTask": 5,
"maxWaves": 50
}
}
}| Setting | Description | Default |
|---|---|---|
concurrency.workers |
Default parallel workers per wave | 2 |
concurrency.max |
Maximum workers allowed (hard ceiling is 10) | 10 |
dependencies |
Dependency scheduling mode: advisory or strict |
"advisory" |
coordination |
Worker coordination level: none, minimal, moderate, chatty |
"chatty" |
messageBudgets |
Max outgoing messages per worker per level (sends rejected after limit) | { none: 0, minimal: 2, moderate: 5, chatty: 10 } |
models.planner |
Model for planner agent | anthropic/claude-opus-4-6 |
models.worker |
Model for workers (overridden by per-task or per-wave model param) |
anthropic/claude-haiku-4-5 |
models.reviewer |
Model for reviewer agent | anthropic/claude-opus-4-6 |
models.analyst |
Model for analyst (plan-sync) agent | anthropic/claude-haiku-4-5 |
thinking.planner |
Thinking level for planner agent | (from frontmatter) |
thinking.worker |
Thinking level for worker agents | (from frontmatter) |
thinking.reviewer |
Thinking level for reviewer agents | (from frontmatter) |
thinking.analyst |
Thinking level for analyst agents | (from frontmatter) |
review.enabled |
Auto-review after task completion | true |
review.maxIterations |
Max review/fix cycles per task | 3 |
planning.maxPasses |
Max planner/reviewer refinement passes | 1 |
work.maxAttemptsPerTask |
Auto-block after N failures | 5 |
work.maxWaves |
Max autonomous waves | 50 |
work.shutdownGracePeriodMs |
Grace period before SIGTERM on abort | 30000 |
work.env |
Environment variables passed to spawned workers | {} |
Each crew agent ships with a default model in its frontmatter. Override any of these via crew.models.<role> in config:
| Agent | Role | Default Model |
|---|---|---|
crew-planner |
planner | anthropic/claude-opus-4-6 |
crew-worker |
worker | anthropic/claude-haiku-4-5 |
crew-reviewer |
reviewer | anthropic/claude-opus-4-6 |
crew-plan-sync |
analyst | anthropic/claude-haiku-4-5 |
Agent definitions live in crew/agents/ within the extension. To customize one for a project, copy it to .pi/messenger/crew/agents/ and edit the frontmatter — project-level agents override extension defaults by name. Agents support thinking: <level> in frontmatter (off, minimal, low, medium, high, xhigh). Config thinking.<role> overrides the frontmatter value.
| Action | Description |
|---|---|
join |
Join the agent mesh |
leave |
Leave the mesh for the current session |
list |
List agents with presence info |
status |
Show your status or crew progress |
whois |
Detailed info about an agent (name required) |
feed |
Show activity feed (limit optional, default: 20) |
set_status |
Set custom status message (message optional — omit to clear) |
send |
Send DM (to + message required) |
broadcast |
Broadcast to all (message required) |
reserve |
Reserve files (paths required, reason optional) |
release |
Release reservations (paths optional — omit to release all) |
rename |
Change your name (name required) |
| Action | Description |
|---|---|
plan |
Create plan from PRD or inline prompt (prd, prompt optional — auto-discovers PRD if omitted, auto-starts workers unless autoWork: false) |
work |
Run ready tasks (autonomous, concurrency optional) |
work.stop |
Stop autonomous work for the current project |
review |
Review implementation (target task ID required) |
task.list |
List all tasks |
task.show |
Show task details (id required) |
task.start |
Start a task (id required) |
task.done |
Complete a task (id required, summary optional) |
task.block |
Block a task (id + reason required) |
task.unblock |
Unblock a task (id required) |
task.ready |
List tasks ready to work |
task.reset |
Reset a task (id required, cascade optional) |
crew.status |
Overall crew status |
crew.validate |
Validate plan dependencies |
crew.agents |
List available crew agents |
crew.install |
Show discovered crew agents and their sources |
crew.uninstall |
Remove stale shared-directory crew agent copies |
| Action | Description |
|---|---|
swarm |
Show swarm task status |
claim |
Claim a task (taskId required) |
unclaim |
Release a claim (taskId required) |
complete |
Complete a task (taskId required) |
Create ~/.pi/agent/pi-messenger.json:
{
"autoRegister": false,
"autoRegisterPaths": ["~/projects/team-collab"],
"scopeToFolder": false,
"nameTheme": "default",
"stuckThreshold": 900,
"stuckNotify": true,
"autoOverlayPlanning": true
}| Setting | Description | Default |
|---|---|---|
autoRegister |
Join mesh on startup | false |
autoRegisterPaths |
Folders where auto-join is enabled (supports * globs) |
[] |
scopeToFolder |
Only see agents in same directory | false |
nameTheme |
Name theme: default, nature, space, minimal, custom |
"default" |
nameWords |
Custom theme words: { adjectives: [...], nouns: [...] } |
— |
feedRetention |
Max events kept in activity feed | 50 |
stuckThreshold |
Seconds of inactivity before stuck detection | 900 |
stuckNotify |
Show notification when a peer appears stuck | true |
autoStatus |
Auto-generate status messages from activity | true |
autoOverlay |
Auto-open overlay when autonomous crew work starts | true |
autoOverlayPlanning |
Auto-open Crew overlay when planning starts or is restored in-progress | true |
crewEventsInFeed |
Include crew task events in activity feed | true |
contextMode |
Context injection level: full, minimal, none |
"full" |
Config priority: project .pi/pi-messenger.json > user ~/.pi/agent/pi-messenger.json > ~/.pi/agent/settings.json "messenger" key > defaults.
Pi-messenger is a pi extension that hooks into the agent lifecycle. It uses pi.on("tool_call") and pi.on("tool_result") to track activity — every edit, commit, and test run gets logged. pi.on("session_start") handles auto-registration, pi.on("session_shutdown") cleans up, and pi.on("agent_end") drives autonomous crew mode by checking for ready tasks after each agent turn.
Incoming messages wake the receiving agent via pi.sendMessage() with triggerTurn: true and deliverAs: "steer", which injects the message as a steering prompt that resumes the agent. File reservations are enforced by returning { block: true } from a tool_call hook on write/edit operations. The /messenger overlay uses ctx.ui.custom() for the chat TUI, and ctx.ui.setStatus() keeps the status bar updated with peer count and unread messages.
Crew workers are spawned as pi --mode json subprocesses with the agent's system prompt, model, and tool restrictions from their .md definitions. Progress is tracked via JSONL streaming — the overlay subscribes to a live progress store that shows each worker's current tool, call count, and token usage in real time. Aborting a work run triggers graceful shutdown: each worker receives an inbox message asking it to stop, followed by a grace period before SIGTERM. The planner and reviewer work the same way — just pi instances with different agent configs.
All coordination is file-based, no daemon required. Shared state (registry, inboxes, swarm claims/completions) lives in ~/.pi/agent/messenger/. Activity feed and crew data are project-scoped under .pi/messenger/ inside your project, so Crew logs live at <project>/.pi/messenger/crew/ and the shared activity feed lives at <project>/.pi/messenger/feed.jsonl. Dead agents are detected via PID checks and cleaned up automatically.
- mcp_agent_mail by @doodlestein — Inspiration for agent-to-agent messaging
- Pi coding agent by @badlogicgames
MIT