self-improvement

Pass

Audited by Gen Agent Trust Hub on Oct 26, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill implements a feedback loop where data from untrusted sources (such as error logs and user-provided corrections) is captured in .learnings/*.md files and promoted to project instructions like CLAUDE.md. This ingestion and promotion process creates a surface for indirect prompt injection if malicious instructions are embedded in the captured data.\n
  • Ingestion points: User feedback, shell error messages, and existing log files in .learnings/ (SKILL.md).\n
  • Boundary markers: Uses Markdown headers and sections (e.g., ### Summary, ### Details) to delimit data.\n
  • Capability inventory: scripts/extract-skill.sh for file creation and instructions to modify project configuration files.\n
  • Sanitization: No explicit sanitization or filtering of the content within the learning entries before promotion.\n- [COMMAND_EXECUTION]: The skill contains shell scripts used for workflow automation.\n
  • scripts/extract-skill.sh scaffolds new skill folders and files. It includes robust regex validation (^[a-z0-9]+(-[a-z0-9]+)*$) for the skill name input to prevent command injection.\n
  • scripts/activator.sh and scripts/error-detector.sh provide diagnostic reminders and fail safely.\n- [EXTERNAL_DOWNLOADS]: Installation documentation (references/installation.md) directs users to download the skill from the vendor's GitHub repository (pskoett/pskoett-ai-skills).
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 26, 2026, 08:53 AM