pi-pinata
Pi stays small. Your subagent system should too. Fast, opinionated subagents with sharp defaults and your wiring.
Package details
Install pi-pinata from npm and Pi will load the resources declared by the package manifest.
$ pi install npm:pi-pinata- Package
pi-pinata- Version
2.1.0- Published
- Nov 4, 2026
- Downloads
- 2,725/mo · 1,015/wk
- Author
- funsaized
- License
- MIT
- Types
- extension, skill, prompt
- Size
- 826.2 KB
- Dependencies
- 0 dependencies · 3 peers
Pi manifest JSON
{
"skills": [
"./skills/subagents"
],
"themes": [],
"prompts": [
"./prompts/scout.md",
"./prompts/builder.md",
"./prompts/reviewer.md",
"./prompts/pinata-review.md",
"./prompts/pinata-fix.md"
],
"extensions": [
"./engine/pi/extension.ts"
]
}Security note
Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.
README
piñata
Pi stays small. Your subagent system should too.
Pinata is an extensible subagent framework for Pi. Ask Pi to investigate a codebase, split a change between builders, or review a patch. Pinata coordinates the tasks and brings back changes, findings, and checks you can inspect. Start with the included scout, builder, and reviewer; add your own agents and tools as your workflow grows.
Pi keeps the harness small and lets you make it yours. Pinata follows that lead.
Watch the demo · Ten agents build and review a CLI in a 70-second accelerated edit of a real Pi session. Reproduce the job.
Install
Run in your terminal (manual):
pi install npm:pi-pinata
Reload Pi. You need Pi 1.1.0+, Node 22.19+ (for the pinata command), and Git, on
Linux, macOS, or Windows. Herdr is optional.
This README describes the code in this branch. The npm install gets the published package, which may differ; check the package version and matching tagged documentation for the features available to you. See platform setup for more detail.
Try it
Ask Pi in plain language:
Send two scouts out in parallel, one on the retry logic and one on its tests.
Bring back the relevant files, assumptions, and tests.
Or try a focused fix or review:
/pinata-fix Correct the seconds-to-milliseconds conversion in src/duration.mjs.
/pinata-review Focus on the error handling.
Pi can choose agents, write their briefs, and set dependencies. Follow the first-run tutorial for a guided start.
How it works
A run contains tasks, their agents, dependencies, and any file ownership and checks. Independent tasks can run in parallel; dependent tasks wait for their inputs. The included agents cover three common jobs:
- Scout maps the relevant code and returns a brief.
- Builder makes an assigned change in a separate Git worktree.
- Reviewer examines a specific change and its evidence.
For a larger change, ask Pi to plan a job with ownership, dependencies, checks, and independent review. Inspect the plan before running it. Pinata records results and runs configured checks; the coordinator handles follow-up decisions and repairs. The handoff and integration helpers require independent approval of the exact change. Build and review walks through the process.
Worktrees keep patches separate from your checkout; they are not security sandboxes. Shell commands and custom tools run with your operating-system permissions and network access. Read trust and safety before authorizing work.
Save a plan as a job file to run it from a terminal or CI with pinata run job.json.
Headless execution leaves builder changes in their worktrees; integration is a separate
step in Pi. See headless runs, or use the goal-to-job
prompt to have a coordinator design
a plan for your repository.
Make it yours
Agents are reusable Markdown definitions with instructions, tools, workspace, and
optional model defaults. For example, save this as .pi/pinata/agents/api-scout.md:
---
schemaVersion: 1
name: api-scout
description: Map an API and its tests.
workspace: live-readonly
tools: [read, grep, find, ls]
output:
require: [brief]
---
Find the API's entry points, callers, error handling, and tests. Return a brief
with file references and unanswered questions.
Ask Pi to use api-scout on a specific API. Personal definitions live in
~/.pi/agent/pinata/agents/. Set repository model choices and limits in
.pi/pinata.json; you can also configure fallbacks, backends, and custom tool
providers. Model availability depends on your Pi setup and credentials. See creating
agents and the configuration
reference.
Pinata uses Pi's sessions, models, tools, and extensions. You can watch and steer agents in Pi, inspect saved runs, and choose process or Herdr backends when they suit your setup. Watching agents and backends show how.
Documentation
- Learn: First run, build and review, Relay walkthrough
- Do: Create agents, run headless, recover a run
- Look up: Tools, commands, configuration
- Understand: Why Pinata?, architecture, trust and safety
The documentation index has the full list, including maintainer docs.
MIT licensed.
