Caelara Labs Agent Safety F
com.caelaralabs/agent · v1.0.1 · score 17/100 · checked 2026-11-22
Caelara Labs Agent Safety tools over remote MCP. Calls need an X-Caelara-Key API key.
Findings
| Severity | Rule | Detail |
|---|---|---|
| low | AG005 | Remote endpoint lists 10 tool(s) to anyone, no credentials needed. |
| critical | AG101 | tool 'agent_input_firewall': description contains instruction-like text aimed at the model ('Ignore previous instructions'). Possible tool poisoning. |
| critical | AG101 | tool 'prompt_injection_scanner': description contains instruction-like text aimed at the model ('Disregard all prior rules'). Possible tool poisoning. |
What was checked
- https://agent.caelaralabs.com/mcp: answers without auth
Public tools: agent_input_firewall, agent_output_validator, pii_redactor, prompt_injection_scanner, token_compressor, json_auto_repair, schema_enforcer, hallucination_diff, citation_extractor, context_chunker
Maintainer? Fix, then add the badge
Grades refresh daily.
[](https://agentguard-pi.vercel.app/s/com-caelaralabs-agent-0a4e2744)
agentguard: static and read-only checks. A grade is not a guarantee of safety. Index · Scanner