Skip to content

Latest commit

 

History

History
139 lines (103 loc) · 3.79 KB

File metadata and controls

139 lines (103 loc) · 3.79 KB

Contributing

Thanks for helping improve Agent Computer.

Agent Computer is an experimental V0 preview: a file-based workspace where coding agents run like apps. Contributions should make the workspace clearer, safer, more useful, or easier to extend.

Good First Contributions

  • Improve documentation for new users.
  • Add public-safe examples.
  • Improve an existing agent's agent.md, workflow.md, or output-template.md.
  • Add focused tests or QA checklists for an agent.
  • Improve local tools under tools/ without adding global install requirements.
  • Propose a community agent package shape for Agent Hub.

Agent App Standard

Every agent app should include:

README.md
agent.md
workflow.md
output-template.md

Executable agents may also include:

tools/
templates/
tests/
examples/
memory/
manifest.json

An agent should not be only a prompt if it needs real capability. Add tools, templates, tests, and validation notes when they make the agent more reliable.

Adding A New Agent

  1. Choose a category:
    • agents/system/ for operating-layer agents.
    • agents/work/ for knowledge-work agents.
    • agents/personal/ for reflective/personal agents.
    • agents/community/ for externally shared agents when that folder exists.
  2. Create the standard files.
  3. Define boundaries and external-action approval rules.
  4. Add tools only when they are local-first and safe to inspect.
  5. Add at least one example or smoke test.
  6. Update system/agent-registry.md.
  7. If routing should discover it, update router logic or docs.
  8. Add a QA note under a project-specific projects/<slug>/qa/ folder while developing, but do not include generated QA artifacts in the public release copy unless they are intentionally public examples.

Tool Rules

Tools must:

  • run inside the workspace
  • avoid global package installs
  • avoid mutating global config
  • fail clearly when dependencies are missing
  • write predictable outputs
  • avoid destructive operations unless explicitly approved

Tools must not:

  • send emails, publish content, make purchases, delete files, change accounts, or access external apps without explicit user approval
  • store secrets
  • write outside the workspace by default
  • silently install dependencies

Documentation Style

  • Be concrete.
  • Separate facts, assumptions, limitations, and recommendations.
  • Prefer examples that show natural-language usage.
  • Do not imply Agent Computer is a stable OS, SaaS product, or finished marketplace.
  • Label V0 limitations honestly.

Public-Safety Requirements

Do not commit:

  • private memory
  • real contacts
  • API keys, tokens, cookies, or credentials
  • private reports
  • converted private documents
  • generated project outputs
  • .DS_Store
  • .env files

Run the release builder before publishing:

npm run release:build -- --version v0-preview --force

Then inspect:

dist/agent-computer-v0-preview/RELEASE_MANIFEST.md

Pull Request Checklist

  • The change keeps Agent Computer local-first.
  • No private/generated artifacts are included.
  • New or changed agents have clear boundaries.
  • External actions remain approval-gated.
  • Relevant docs are updated.
  • Focused smoke or QA checks were run.
  • Known limitations are stated.

Agent Hub Contributions

Agent Hub is a future community sharing platform, not a paid marketplace in V0.

For proposed community agents, include:

  • a short description
  • agent category
  • manifest draft
  • permissions needed
  • local tools included
  • example input/output
  • safety notes

See Agent Hub.

Sources

This contribution guide is based on: