Human Writing Rules processes source material, prompts, generated artifacts, adapter packets, and optional provider credentials. Treat every external adapter and source snapshot as a trust-boundary crossing.

Security fixes are provided for the latest stable 1.x release. A release-candidate or historical draft receives a fix only when the issue also affects the supported stable line or blocks stabilization.

Use GitHub private vulnerability reporting for dev-ik/human-writing-rules. Do not open a public issue containing an exploit, credential, private source, personal data, unpublished communication, or proprietary benchmark input.

Include:

• affected version and commit when known;
• affected command, schema, adapter, or workflow;
• reproduction steps using synthetic data;
• expected and observed security boundary;
• impact and whether network or live-provider access is required.

If private vulnerability reporting is unavailable, open a public issue without sensitive details and request a private contact channel.

The maintainer should acknowledge a report within seven days, confirm scope before publishing details, and coordinate a fix and disclosure date with the reporter. These are project targets, not a service-level agreement.

• Core validation and release verification are offline by default.
• Provider credentials belong in environment variables and must not enter tasks, fixtures, logs, release manifests, or adapter results.
• Source snapshots reject unsafe paths, symlink escapes, oversized files, and digest mismatches.
• Subprocess adapters are trusted code but receive a bounded packet and shell-free invocation.
• Generated visuals, prompts, source material, and benchmark inputs retain their privacy, license, likeness, and disclosure obligations.

Never include real secrets or private data in a proof of concept.