improve-system

Pass

Audited by Gen Agent Trust Hub on Jul 18, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a local Python script located at ~/.claude/hooks/maintenance_due.py to mark completion of the maintenance cadence.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8) by processing untrusted data from various sources to derive system rules.
  • Ingestion points: Processes content from agent-improvements/ logs, session memories in .remember/, and feedback_* files which may contain untrusted data from prior interactions.
  • Boundary markers: No specific boundary markers or instructions to ignore embedded commands within the logs are present.
  • Capability inventory: Uses the Edit tool to modify core system files (e.g., CLAUDE.md, skill files, agent configurations) and executes shell commands via Python.
  • Sanitization: No sanitization or validation of the ingested log content is performed before clustering patterns and drafting proposals.
  • Risk: Malicious content embedded in logs from a previous session could be misinterpreted as a valid behavioral pattern, potentially tricking the agent into proposing harmful system-level modifications.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 18, 2026, 11:09 AM
Security Audit — agent-trust-hub — improve-system