If you believe you have found a security vulnerability in relevo, please report it privately. Do not open a public GitHub issue or pull request that describes the problem before a fix is available.

Preferred channel:

• Open a private security advisory: https://github.com/davo20019/relevo/security/advisories/new

Alternative channel:

• Email the maintainer at davo20019@gmail.com with subject relevo security report.

When reporting, please include:

• A description of the issue and the impact you believe it has.
• Steps to reproduce, including a minimal agents.json or command sequence when relevant.
• The relevo version (relevo --version or the installed package version) and your operating system and Node.js version.
• Any proof-of-concept code, logs, or screenshots that help us reproduce it.

You can expect:

• An acknowledgement within 3 business days.
• An initial assessment within 7 business days.
• A coordinated disclosure timeline once the impact is understood. We aim to ship a fix and publish a GitHub Security Advisory before any public details are shared.

relevo is in active 0.x development. Security fixes are published against the latest minor release on npm.

relevo is a local orchestration layer for coding-agent CLIs that the user has already installed and authenticated. It does not make outbound network calls of its own; it spawns child processes that do.

In scope for this policy:

• Command, argument, or path injection in how relevo invokes child CLIs.
• Path traversal in task names, agent keys, transcript paths, or other filesystem operations relevo performs.
• Unsafe handling of agents.json (built-in, user, or project-local).
• Information disclosure in transcripts, session files, or other state written by relevo.
• Supply-chain issues in the published relevo npm package itself.

Out of scope (please report to the upstream project instead):

• Vulnerabilities in Claude Code, Codex, Cursor, OpenCode, Antigravity (agy), or any other underlying agent CLI.
• Issues that require an attacker who already has interactive shell access as the user running relevo.
• The intentional automation defaults documented in README.md (skipping permission prompts, sandbox-off, MCP auto-approval, etc.). These are documented trade-offs of running headless coding agents in parallel and can be tightened in agents.json.

Thank you for helping keep relevo and its users safe.