New:Microsoft Teams Notifications Are Now Available in Socket.Learn more →
Get Started

@bosun-sh/chest

Package Overview
Dependencies
Maintainers
1
Versions
3
Alerts
File Explorer

Advanced tools

Socket logo

Install Socket

Detect and block malicious and high-risk dependencies

Install

@bosun-sh/chest

Local governance store for AI agent workflows.

latest
Source
npmnpm
Version
0.2.0
Version published
Weekly downloads
7
250%
Maintainers
1
Weekly downloads
 
Created
Source

Chest

CI npm License

Chest is an open-source self-governance store for AI agent workflows.

It keeps project governance criteria in local, versioned files so agents can plan, execute, and validate work against explicit goals without turning every decision into a human checkpoint.

Install

bun add @bosun-sh/chest

The package is published as a normal npm artifact and exposes:

  • @bosun-sh/chest for the core registry and tool IDs
  • @bosun-sh/chest/ohtools for the default prewired Ohtools app
  • @bosun-sh/chest/cli for the CLI-attached app
  • @bosun-sh/chest/mcp for the stdio MCP app

Use

If you want to embed Chest in your own launcher, import the registry or one of the ready-made app entrypoints:

import { createChestApp, chestCliApp, chestMcpApp, chestToolIds } from "@bosun-sh/chest";
import chestApp from "@bosun-sh/chest/ohtools";

console.log(chestToolIds.length);
console.log(chestApp.build().adapters.has("mcp"));
console.log(chestCliApp.build().adapters.has("cli"));
console.log(chestMcpApp.build().adapters.has("mcp"));

const registry = createChestApp().build();
console.log([...registry.tools.keys()]);

To launch Chest through ohtools, create a tiny local wrapper and point the CLI at it:

// chest-app.ts
import chestApp from "@bosun-sh/chest/ohtools";

export default chestApp;
bunx ohtools --app ./chest-app.ts list

What It Solves

AI-assisted workflows often need governance, but repeated checkpoints can turn into a throughput bottleneck. Chest is designed for projects where the agent can evaluate most decisions against machine-verifiable criteria instead of stopping for approval every time.

Chest organizes that governance around:

  • OKRs
  • KPIs
  • Definition of Ready
  • Definition of Done

The project is based on the intervention-penalty problem described in: The Intervention Penalty: A Simulation Study of Human Checkpoint Costs in AI Coding Governance.

Storage

Chest stores its local state under .chest/ by default:

  • .chest/okrs.yaml
  • .chest/kpis.yaml
  • .chest/definition-of-ready.yaml
  • .chest/definition-of-done.yaml
  • .chest/reports/*.yaml

Set CHEST_HOME to redirect the store to another location.

Current State

Chest is implemented as a real @bosun-sh/ohtools registry with 20 Chest tool IDs exposed through the CLI and MCP surfaces. The checked-in implementation includes:

  • src/ohtools.ts as the registry composition root used by bunx ohtools.
  • src/workspace/cli-adapter.ts for CLI attachment.
  • src/workspace/mcp-server.ts as the stdio MCP bridge used by the e2e parity test.
  • YAML-backed OKR, KPI, DoR, DoD, validation, judge, and report workflows.
  • Validation and report tools that persist report files under .chest/reports/.

Development

The repo uses the same commands locally and in CI:

bun run lint
bun run test:unit
bun run test:e2e
bun run build
bun run pack:check
bun run smoke:packed

Git hooks are managed with Husky:

  • pre-commit runs lint and test:unit
  • pre-push runs test:e2e and build

CLI error envelopes are JSON and exit nonzero; capture the error stream when parsing failures such as OHTOOLS_VALIDATION_ERROR.

Repository Layout

  • src/ contains the Chest registry, adapters, domain slices, ports, and workspace helpers.
  • tests/ contains unit, integration, and e2e coverage.
  • docs/ contains the contract specs for the implemented v1 feature slices.
  • .github/workflows/ci.yml defines the GitHub Actions pipeline.

Contributing

Issues and pull requests are welcome.

If you plan to change behavior, keep the docs, tests, and public tool surface aligned with the implementation. The repo is intended to stay easy for agents and humans to navigate.

License

Apache-2.0. See LICENSE and package.json for the current license declaration.

Keywords

ai

FAQs

Package last updated on 28 Jun 2026

Related posts