agent-evidence is a narrow local package for runtime evidence export, operation-accountability profile validation, offline bundle verification, verification receipts, signed export metadata, and reviewer-facing review packs.

Security review covers:

• the current main branch
• the v0.6.x release line

Older release lines may be useful for historical reproducibility, but they are not the active security-support target.

In-scope security areas are:

• local evidence packaging
• profile validation
• offline bundle verification
• signed export metadata
• verification receipts
• reviewer-facing review packs

Out-of-scope areas include hosted verification services, remote attestation, legal non-repudiation, compliance certification, and production forensic timestamping.

Report suspected vulnerabilities through GitHub security advisories when available, or open a GitHub issue with a minimal non-sensitive reproduction.

Do not include private keys, tokens, credentials, production secrets, sensitive runtime evidence, or private customer data in reports, issues, pull requests, logs, or test fixtures.

Useful reports include:

• affected command, file, or package surface
• expected and actual behavior
• minimal redacted input needed to reproduce the issue
• whether the issue affects validation, export, bundle verification, signed metadata, or review-pack generation

Maintainers and contributors must not commit:

• private keys
• tokens
• credentials
• production secrets
• sensitive runtime evidence
• unredacted private logs or review packs

Signed export tests and examples should use test fixtures only. Do not reuse production signing material in this repository.

Security-sensitive pull requests should be reviewed for:

• signed export metadata
• manifest and hash verification
• bundle path traversal risks
• untrusted input validation
• JSON, CSV, and XML export safety
• offline verification receipts
• dependency updates
• GitHub Actions workflow permissions

This project does not claim:

• legal non-repudiation
• court-grade audit proof
• regulatory certification
• AI Act approval
• production forensic timestamping
• official FDO standard status
• hosted verification service
• default sandboxing or access control