hermes-insights

Pass

Audited by Gen Agent Trust Hub on Jun 23, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill accesses sensitive local directories including ~/.claude/projects/ and ~/.claude/skills/. These paths contain historical session data, decisions, learned facts, and project-specific metadata. While no network exfiltration patterns (such as curl or fetch) were detected, the skill requires read access to the user's full interaction history with the agent.
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands such as ls and find to crawl the local filesystem and identify session memory files for analysis.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface because it processes untrusted data from external session memory files.
  • Ingestion points: Reads files at ~/.claude/projects/*/memory/session_*.md.
  • Boundary markers: The instructions do not specify the use of delimiters or warnings to ignore embedded instructions within the processed logs.
  • Capability inventory: Uses ls and find to locate data; analysis is performed within the agent context.
  • Sanitization: No explicit sanitization or filtering of the content from memory files is mentioned before processing.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 23, 2026, 05:23 PM
Security Audit — agent-trust-hub — hermes-insights