Clear rules for an open security list.
threat.cafe publishes an open, human-reviewed list of compromised agent skills, MCP servers, and packages. Using our feed, local hook, or hosted decision log means agreeing to these plain-language terms.
License & verdict summary
The public list stays open under CC-BY 4.0 and the local hook is Apache-2.0. We never label any package, skill, or MCP server as safe.
Retain attribution to threat.cafe and the credited finder.
What you can do with the list and the hook.
We keep the core dataset open so every solo developer and security researcher can guard their agents without a paywall.
Three-state verdict contract.
Every lookup returns known bad, no known bad, or unknown. We never label any artifact as safe. A verdict of no known bad means only that the specific build has no active compromise record on our reviewed list at the time of check.
CC-BY 4.0 open list license.
You are free to copy, mirror, and adapt /feed.json for personal or internal company use provided you retain attribution to threat.cafe and the credited upstream security researchers on each record.
Apache-2.0 local hook software.
The local PreToolUse hook script (/hook/threat-cafe-hook) is licensed under Apache 2.0 on an as-is basis. You control your local policy.yml and remain responsible for reviewing actions taken by your coding agents.
Commercial gateway & MSP embedding.
Reselling or embedding the signed feed inside a commercial multi-tenant agent gateway, managed security service, or commercial IDE product requires a Team, MSP, or Platform agreement.
Evidence standards and limitation of liability.
Our goal is accurate, source-backed operational intelligence with fast remediation when upstream maintainers ship clean builds.
Legal & operational terms
-
Evidence
Primary technical proof. Every public record cites reproducible indicators of compromise, NVD/CISA advisories, or primary security research write-ups.
-
Disputes
48-hour correction path. Maintainers who remediate a compromised release or dispute a listing receive human review within 48 hours under our curation policy.
-
As-is basis
No warranty of completeness. New supply-chain attacks appear daily. An unknown or no known bad verdict does not warrant that third-party code is free of defects or backdoors.
-
Liability
Limitation of liability. To the maximum extent permitted by law, threat.cafe, manav.id, and TheWORKCompany shall not be liable for indirect, incidental, or consequential damages arising from agent tool execution.
These terms are governed by applicable law and may be updated as commercial Team and MSP features launch.
Acceptable use of /api/ endpoints
- /api/intake.php: Use only for genuine early access signups, good-faith security finding reports, and verifiable maintainer disputes. Automated scraping or spam flooding is rate-limited.
- /api/v1/events.php: Use to record structured agent hook verdicts for your authorized workspace. Do not submit raw credentials, PII, or secret keys in artifact strings.
- /api/v1/audit-pack.php: Export JSON and CSV compliance attestations for internal governance, SOC 2, and ISO 42001 audits.
Attribution checklist (CC-BY 4.0)
- Credit threat.cafe as the feed curator.
- Preserve the credited_finder field on every record.
- Link back to https://threat.cafe/feed.json.
- Never relabel no_known_bad or unknown records as "safe".
Operator & legal notices. threat.cafe is part of manav.id (TheWORKCompany). Questions regarding licensing, commercial gateway agreements, or disputes: [email protected].