run-mcpjam-evals
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [SAFE]: No malicious patterns detected. The skill provides legitimate instructions for interacting with the MCPJam evaluation platform.
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing external data from the evaluation environment.
- Ingestion points: Evaluation results are retrieved via get_eval_run_steps (evidence URLs, reason strings) and list_eval_run_iterations (actual tool calls) as described in SKILL.md and references/triage.md.
- Boundary markers: No explicit instruction delimiters are mentioned for the processing of external content.
- Capability inventory: The skill can execute run_eval_suite (financial cost), create_eval_suite (state modification), and generate_eval_cases (AI generation) via the connected MCP server.
- Sanitization: No sanitization of retrieved evaluation evidence, log data, or reason strings is performed before the agent analyzes it.
- [EXTERNAL_DOWNLOADS]: The skill references official vendor resources, such as @mcpjam/sdk and mcpjam-eval-import, which are documented for use in local authoring and import workflows.
Audit Metadata