Do not open a public issue containing exploit details, credentials, private repository content, or customer data. Report vulnerabilities privately through the repository host's security-advisory feature. Maintainers should acknowledge the report within five business days and coordinate disclosure after a fix is available.

• Keep patchlog.yaml local and use environment-variable references. The file is ignored; copy patchlog.example.yaml to start.
• Never commit API tokens. If a token appears in Git history, terminal logs, CI output, or a shared review, revoke and replace it immediately; deleting the current file is not sufficient.
• Credential-bearing integrations require HTTPS. Loopback HTTP is allowed for local test servers; any other exception requires the explicit security.allow_insecure_credentials setting.

When an AI feature is enabled, patchlog may send commit messages, selected code diffs, linked issue text, and computed release metrics to the configured AI endpoint. Before the first remote request, the CLI prints the endpoint, data categories, excluded file patterns, and request-size limit. A shared boundary redacts common secret formats, excludes configured sensitive/generated paths, and rejects prompts over ai.max_input_chars. Redaction is defense in depth; review endpoint ownership and exclusions before sending private source code.

Official release assets include SHA256SUMS. Verify the checksum before installing and prefer an immutable version instead of latest.