Skip to content

Repository files navigation

oy

Crates.io docs.rs

Deterministic-input audits, code reviews, and one-finding fixes — as portable agent skills.

oy helps your coding agent review a repository without quietly choosing a small sample. It prepares an ordered, reviewable set of files, lets your agent's model analyze them under its own permissions, and verifies the report before writing it. The workflows are standard Agent Skills that OpenCode, Cursor, Codex, Copilot, and Gemini CLI all discover from .agents/skills.

What you get

  • oy-audit skill — security-focused repository audits (ISSUES.md or SARIF)
  • oy-review skill — whole-workspace or target-diff code reviews (REVIEW.md)
  • oy-enhance skill — fix one reported finding at a time
  • oy-setup skill — agent-driven setup, verification, and persona installation
  • deterministic CLI: oy audit|review prepare and finalize, oy setup, oy doctor

Your agent owns models, credentials, sessions, and general tools. oy adds the evidence and report workflow; it is not a second agent runtime or permission system.

Quick start — 2 minutes

You need: Linux or macOS (WSL2 on Windows), any agent that reads Agent Skills (OpenCode, Cursor, Codex, Copilot, or Gemini CLI) with a model provider already configured.

Step 1 — Install oy:

curl -fsSL https://oy.adonm.dev/install.sh | sh
# choose Global when prompted (or pass --global / --workspace to skip the prompt)
oy doctor --check   # expect: "global skills ok" or "workspace skills ok"

Review install.sh before piping to a shell. Prefer a manual install? See Getting started.

Step 2 — Ask your agent (copy-paste one line):

run the oy-setup skill to finish setup

Then create your first report:

audit this repository with the oy-audit skill

That's it — look for ISSUES.md in your workspace root. Try next:

review the diff against main with the oy-review skill
use the oy-enhance skill to fix audit-0123456789abcdef

Local dev from this checkout: just install (cargo install + oy setup + oy doctor --check).

What just happened?

  1. oy collected eligible files into ordered chunks under .oy/runs/<id>/
  2. Your agent read every chunk and wrote candidate findings
  3. oy verified the evidence wasn't changed and normalized the report

The inputs are deterministic; the conclusions are not. Model choice still affects findings. See Coverage and limits before using a report for high-assurance work.

New to Agent Skills?

Agent Skills are just SKILL.md files. oy setup writes them to ~/.agents/skills/ (or .agents/skills/ for workspace-only). Your agent discovers them automatically — no API keys or separate daemon. If oy doctor --check passes but your agent can't see the skills, ask it to run the oy-setup skill; it will copy/symlink them to your host's preferred location (e.g. .claude/skills).

Common workflows

Audit a repository

audit this repository with the oy-audit skill
audit src/auth with the oy-audit skill
audit the authentication boundaries with the oy-audit skill
audit with sarif output and write oy.sarif

A single existing workspace path narrows collection (e.g. src/auth). Other text is treated as review guidance for the model.

Review code

review this repository with the oy-review skill
review the diff against main with the oy-review skill
review the diff against main with the oy-review skill, focusing on error handling

A branch, commit, tag, or ref selects target-diff review. Without a target, the skill reviews the workspace.

Fix one finding

use the oy-enhance skill to fix audit-0123456789abcdef

Reports include stable finding IDs. The skill confirms the cited source, makes one focused fix, and runs the narrowest available verification. Rerun the originating audit or review to confirm.

Troubleshooting

  • oy: command not found → restart your shell (mise activation) or check ~/.local/bin is on PATH
  • oy doctor --check fails → run oy setup again, then ask your agent to run the oy-setup skill
  • Agent can't find the skill → see Compatibility for where each agent looks, or ask the oy-setup skill to copy them
  • exceeds max-chunks 80 → narrow the path first (e.g. audit src/auth), only then raise --max-chunks
  • Model not configured → configure your provider in your agent (not in oy); oy never stores credentials

More help: Getting started · Workflow guide · Troubleshooting · oy doctor · oy <command> --help

Safety

oy is not a sandbox. Prepared source may be sent to your configured model provider. The skills run under your agent's own permissions. Use a disposable environment for untrusted repositories and read SECURITY.md.

Documentation

Run oy <command> --help for the installed version's exact flags.

About

A focused OpenCode agent with repeatable audits, code reviews, and one-finding fixes.

Topics

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages