Caliper is pre-1.0; only the latest released minor version receives security fixes.
Please report security issues privately via GitHub Security Advisories ("Report a vulnerability" on the repository's Security tab). Do not open public issues for vulnerabilities.
Include: affected version, a minimal reproduction, and impact. Never include real session data, credentials, or private repository content — the synthetic demo dataset (caliper demo, caliper experiment demo) reproduces all rule and pipeline paths for safe reproduction.
We aim to acknowledge reports within 7 days.
Full details in docs/PRIVACY.md. Key guarantees, enforced by tests:
- Local-only analysis. No network calls, telemetry, or LLM during audit. Experiments run agents locally; paid inference is explicitly approval-gated (
--approve-inference). - Read-only sources. Agent databases/files are opened read-only; byte-for-byte immutability is tested.
- Content-free outputs. Reports/registry contain metrics, fingerprints, paths, and categories — never conversation text, tool output, raw tool input, or credentials. Credential stores (OpenCode credential tables, Claude
~/.claude.json/.credentials.json, Codexauth.json) are never opened. - Sanitization. Secret redaction before truncation; control-character sanitization before terminal rendering; parser exception messages never propagated.
- Isolated experiments. Fresh workspace per trial; hard timeouts with process-group termination; transient execution artifacts deleted; atomic run locks.
- State permissions. Caliper's state directory is written
0600, atomically.
See the "Known limitations" section of docs/ARCHITECTURE.md and docs/REVIEW-STATUS.md for the current review status and disclosed residual risks (e.g., session titles are user-derived free text; Codex/Claude live execution backends are unverified and gated).