ClarityIT takes security reports seriously. Please do not file public GitHub issues for security vulnerabilities.
Instead, report suspected vulnerabilities privately to the maintainers. Include:
- A description of the issue and its potential impact
- Steps to reproduce (proof of concept if possible)
- The affected version/commit
You should receive an acknowledgement within 72 hours. We will coordinate a fix and disclosure timeline with you.
ClarityIT is designed around a sovereign-hybrid deployment model with an explicit agent autonomy boundary:
- Agent autonomy is capped at A4. A5 (full autonomy) is hardcoded-disabled — rejected before any database lookup and excluded by a
CHECKconstraint. - The reasoning worker is isolated. It has no
DATABASE_URL,NATS_URL,REDIS_URL, orMINIO_ENDPOINT. It fails closed at startup if any are set and communicates only over HTTP to the Go API. - No destructive mutations via agents. Agents emit intentions through a 13-check Tool Gateway policy chain; allowed mutations are limited to start/shutdown/stop/snapshot.
- High-risk operations require MFA + approval + policy + audit + outbox before execution.
chain_of_thoughtis always stripped from model output; onlyreasoning_summaryis retained.
See docs/security/agent-autonomy-boundary.md and docs/architecture/ for the full threat model.
This policy covers the ClarityIT source repository. It does not cover third-party dependencies (report those upstream) or deployments you do not control.