Qarinah is a local project-memory library and CLI. It is designed to make captured records tamper-evident and retrieval bounded; it is not an operating-system sandbox, credential vault, or complete enforcement boundary.
| Version | Security fixes |
|---|---|
Current 0.1.x release |
Supported |
| Older prereleases and unreleased source snapshots | Upgrade required |
Confirm the current public version with npm view qarinah version.
Use the repository's private vulnerability reporting flow. Include:
- the affected version and operating system;
- the exact command, API, hook, or MCP method;
- a minimal reproduction using synthetic data;
- expected and observed behavior;
- impact, prerequisites, and suggested mitigations if known.
Do not put credentials, private project context, private transcripts, exploit details, or undisclosed vulnerabilities in a public issue. If private reporting is unavailable, open a public issue containing no exploit detail and ask the maintainer to establish a private channel.
The maintainer will acknowledge a complete report, reproduce it, coordinate a fix and disclosure, and credit the reporter unless anonymity is requested. Response time is best effort; the open-source project does not promise a service-level agreement or bug bounty.
Reports are especially useful when they demonstrate:
- capture without explicit machine-local trust;
- workspace escape, path traversal, or linked-file substitution;
- secret exposure that bypasses documented bounds and redaction;
- event-log, checkpoint, or derived-state integrity failures;
- authority expansion through MCP, Codex, Claude Code, Maqam, or interoperability adapters;
- prompt or retrieved-content execution by Qarinah;
- denial of service that bypasses documented input and resource ceilings;
- supply-chain compromise in the packed npm artifact or release workflow.
Claims that a retained statement is factually true, model-provider behavior, and unregistered host side effects are outside Qarinah's security guarantee unless Qarinah itself violates a documented boundary.
The detailed threat model, defaults, non-goals, and known limits are maintained in docs/SECURITY.md. Privacy behavior is documented in PRIVACY.md.