Skip to content

Security: plmn95/punch

SECURITY.md

Security policy

Reporting a vulnerability

Please do not open a public issue for a suspected vulnerability. Use GitHub's private vulnerability reporting for this repository. Include the affected version, reproduction conditions, impact and a minimal proof that contains no real credentials or customer data.

You should receive an acknowledgement within seven days. No bounty programme or response-time guarantee is currently offered.

Supported version

Security fixes target the latest 0.1.x release and the current main branch.

Security model

Punch accepts untrusted public webpages and sends bounded, normalised evidence to Anthropic. Its network layer rejects non-public addresses, credential-bearing URLs, unsafe redirects, unsupported response types and responses outside fixed budgets. Generated semantic data is Zod-validated and then checked against source evidence before rendering.

API keys are accepted by the Anthropic adapter at construction or read by the CLI from ANTHROPIC_API_KEY. They are not ordinary campaign input and must not be placed in URLs, instructions, traces or committed files.

Traces can contain normalised business and product information. They are opt-in and should be retained only as long as needed. Punch does not archive raw pages or raw provider requests and responses in output artifacts.

Deployment responsibility

Punch generates files and does not send email. Operators remain responsible for reviewing generated content, replacing compliance placeholders, protecting output files and credentials, and meeting the requirements of their provider, destination and jurisdiction.

There aren't any published security advisories