self-improvement

Pass

Audited by Gen Agent Trust Hub on Mar 8, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill includes several shell scripts (scripts/activator.sh, scripts/error-detector.sh, scripts/extract-skill.sh) designed to automate learning reminders and facilitate skill scaffolding.
  • [COMMAND_EXECUTION]: scripts/extract-skill.sh performs file system operations, including directory creation (mkdir -p) and template-based file generation. It implements validation logic to prevent the use of absolute paths or directory traversal segments (..) in output paths.
  • [PROMPT_INJECTION]: The skill implements a feedback loop that creates a surface for indirect prompt injection. It captures untrusted data (user corrections, tool error messages) and stores them in .learnings/*.md files, which are intended to influence the agent's behavior in future sessions.
  • Ingestion points: User input (corrections/feature requests) and tool output (captured via the CLAUDE_TOOL_OUTPUT environment variable in error-detector.sh).
  • Boundary markers: The skill uses structured Markdown templates with headers (e.g., ### Summary) and metadata fields to delimit content.
  • Capability inventory: The skill includes scripts for file system manipulation and is intended to run in environments with CLI tool access.
  • Sanitization: No explicit sanitization or escaping of captured content is performed before logging to the markdown files.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 8, 2026, 10:21 AM