Optional agent skills
Make the safe first move.Before code runs.
The repyy agent skill gives a coding agent a repeatable instruction set for preflighting unfamiliar repositories. It asks the agent to scan first, preserve the target as data, explain the evidence, and report unresolved coverage. When the CLI is unavailable, it continues with a limited static review.
The skill is guidance for an agent. Scanner results require a trusted repyy executable; manual review remains available without one.
Install the skill
Run one of these commands in your terminal. The interactive form lets the skills tool prompt you; the global form installs instructions for Codex without prompts.
npx skills add Kevin-Umali/repyy --skill repyynpx skills add Kevin-Umali/repyy --skill repyy -g -a codex -yInstall Node.js and npm so npx is available, install Git and keep it on PATH, and allow network access to fetch the published skill. Git and network are needed for this skill installation, even though a local repyy scan needs neither.
Neither command installs dependencies from a repository you plan to scan. To use the scanner, install repyy separately using theinstallation guide, then confirm it with repyy version.
Install a helper from the same repository with npx skills add Kevin-Umali/repyy --skill NAME, replacingNAME with the skill you need. The main skill recommends helpers when useful and continues if one is missing.
All four skills use the host's built-in question tool when available to ask for missing targets, reports, or goals. They use inputs already supplied and do not install missing tools or execute the target.
Scanner targets are directories or supported Git URLs. For an individual artifact, the skill stages only that approved regular file in a private directory and scans the directory without extracting or executing it. It does not expand the review to a broad parent folder. A single skill file leaves missing scripts and references outside the review.
What the skill tells an agent to do
- Use the target and goal already supplied; ask for missing inputs with the built-in question tool when available.
- Check the trusted installed scanner, intelligence snapshot age, and rule provenance offline; use limited manual review if the CLI is unavailable.
- Scan every requested target together where possible and save bounded JSON.
- Render a local HTML report from that JSON without rescanning source.
- Validate important findings against surrounding source and trace suspicious inputs toward execution or data destinations.
- Report coverage, findings, locations, context, confidence, counterevidence and remediation without granting execution permission.
- Recommend a report, manual-review, or intelligence helper when useful, and continue without installing it if it is unavailable.
repyy scan --format json --output repyy.scan.json TARGET...repyy report repyy.scan.json --format html --output repyy.report.htmlThe agent should treat SCAN INCOMPLETE as unresolved and state that NO FINDINGS is not proof of safety. A manual review must not invent a scanner verdict. It should not paste credential-shaped evidence into chat.
Use the opt-in Docker backend
The skill includes an opt-in Docker path for local folders and HTTPS remotes. Prepare Docker and the exact signed, digest-pinned image for the installed repyy release before asking the agent to scan.
repyy scan TARGET --sandbox=docker --format html --output repyy.report.htmlDocker rejects SSH URLs and --keep-workdir. Missing Docker or a missing release-matched image is a preflight error; the skill must never accept a silent host fallback. Read the isolation guide for Docker setup and image verification.
Understand the limits
repyy intel update unless the user asks.The skill is a workflow guardrail, not a guarantee that a repository is safe. A human still decides whether to execute code, install dependencies, or continue in an isolated environment.
Therepository copy of the skill ↗is the canonical reference for agent behavior. Keep it aligned with the installed CLI and review it when upgrading repyy.