jev-compaction
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute specific shell commands to export and process session data.\n
- Evidence: Use of
hermes sessions export --format jsonl -to retrieve session history andjev compact-select --digest < transcript.jsonto process the data.\n- [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves processing conversation transcripts, which are untrusted sources that could contain malicious instructions.\n - Ingestion points: Exported Hermes session history in JSONL format (SKILL.md).\n
- Boundary markers: The skill uses a structured digest output with
[KEEP VERBATIM]and[background]tags to delineate data types for the agent.\n - Capability inventory: Shell execution capabilities and natural language generation for handoffs.\n
- Sanitization: The skill documentation explicitly states that turns appearing to contain secrets are not sent to the judging logic and default to being summarized, providing a heuristic safety layer.
Audit Metadata