continuous-learning

Pass

Audited by Gen Agent Trust Hub on Jun 24, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes a bash script (evaluate-session.sh) to read session transcript metadata and determine if the session meets the criteria for pattern extraction. This script performs basic file system operations such as directory creation (mkdir -p) within the user's home directory for storage of learned skills.
  • [COMMAND_EXECUTION]: The documentation provides instructions to configure a 'Stop' hook in the platform settings (~/.claude/settings.json). This allows the evaluation script to execute automatically at the conclusion of every session.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes session transcripts which can contain untrusted data encountered during the session.
  • Ingestion points: The evaluate-session.sh script reads the transcript path provided by the agent environment.
  • Boundary markers: No specific delimiters are used to separate transcript content from instructions during the extraction phase.
  • Capability inventory: The skill's primary capability is the creation and management of local skill files based on processed session data.
  • Sanitization: No sanitization or filtering is applied to the transcript content before the agent is asked to evaluate it for patterns.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 24, 2026, 12:57 PM
Security Audit — agent-trust-hub — continuous-learning