ClawBio runs bioinformatics skills on a user's own machine, often against that user's own genome. A vulnerability here can expose genetic data, so reports are handled privately until a fix or mitigation is available.

Do not open a public issue for a security problem.

1. Preferred: GitHub private vulnerability reporting. Open https://github.com/ClawBio/ClawBio/security/advisories/new and describe the problem. The report is visible only to the maintainers until an advisory is published.
2. Fallback: email the maintainer listed in pyproject.toml (mc.admin@manuelcorpas.com) with ClawBio security in the subject line.

Include what you can of: the skill or module affected, the version or commit, a minimal reproduction, and what an attacker gains. A proof of concept against the bundled demo data is welcome. Never send real patient or personal genomic data in a report.

ClawBio has one maintainer at present, so these are commitments made in good faith, not contractual service levels.

• Acknowledgement within 5 working days.
• An initial assessment (accepted, needs more information, or out of scope) within 14 days.
• A fix or a documented mitigation targeted within 90 days of acceptance, and sooner for anything that exposes genetic data or executes untrusted input.
• Coordinated disclosure: a public advisory and a CHANGELOG entry are published with the fix. Reporters are credited unless they ask not to be.

In scope:

• The clawbio Python package, the CLI, and every skill under skills/.
• Skill documentation, demo data or fixtures that could steer an AI agent into unsafe behaviour (prompt injection through SKILL.md or bundled files).
• Secrets, tokens or credentials committed to the repository, or leaked through a skill's output or reproducibility bundle.
• Supply chain: dependency pins, uv.lock, GitHub Actions workflows, and the PyPI release path.
• Any skill that sends data off the machine without saying so in its SKILL.md and in docs/data-handling.md.

Out of scope, please use a normal issue instead:

• Scientific accuracy: wrong annotations, misleading percentiles, incorrect guideline mappings. These matter and are handled in the open through issues and ClawBench, not through this policy.
• Vulnerabilities in third-party services that a skill calls (Ensembl, NCBI, PGS Catalog, Open Targets and others). Report those upstream, and tell us if ClawBio should stop calling the service.
• Rate-limit or denial-of-service effects on public APIs caused by running a skill in a loop.

Security fixes go into the latest minor release only.

Research that follows this policy, avoids accessing or destroying other people's data, and gives the maintainers reasonable time to respond will be treated as authorised and welcome.