The latest release line (0.1.x) receives fixes. Pre-1.0 the API may change between minor versions.

Please use GitHub's private vulnerability reporting on this repository (Security → Report a vulnerability). Do not open a public issue for anything that could leak a secret or expose a user.

Include: affected version, a description of the issue, a minimal reproduction (use fake secrets only), and the impact you expect. We aim to acknowledge within a few days.

maskgap is a tool that reads live secrets by design — you point it at the values you want protected. Its safety properties are about what it does with them:

• Output discipline. Findings and reports carry a label and a non-reversible sha256: fingerprint, never the value. Preview lines have every known match masked before they are printed. The CLI test suite asserts this end to end.
• No side channels. No network access, no telemetry, no update checks, no subprocesses, no eval, no deserialisation of untrusted data, no temporary files holding secrets except the atomic-rewrite temp file of --in-place (same directory, renamed over the original).
• Zero runtime dependencies. The supply-chain surface is the Python standard library plus this repository's code. Dev tooling is pinned exactly and audited with pip-audit on every push (see docs/security-audit.md).
• Read-only unless asked. scan, audit and encoders never modify input; redact writes to stdout or --output, and only --in-place rewrites files.

• Secrets that were never loaded into the scanner cannot be found by value matching (provider-shaped tokens are covered heuristically; see the README).
• A compromised host, shell history, or process list is outside the threat model — maskgap protects what you feed it, not the environment.
• The coverage model reflects the public actions/runner source and can lag the platform; gaps in that model are documentation bugs, not vulnerabilities. Please open a regular issue for those.