skill-auto-improver

Pass

Audited by Gen Agent Trust Hub on May 16, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted instruction files (SKILL.md and referenced markdown files) from a target directory to perform automated edits. This creates a surface for indirect prompt injection, where a malicious skill being 'improved' could contain instructions intended to influence or override the agent's behavior during the loop.
  • Ingestion points: Target skill directory (SKILL.md and files in references/)
  • Boundary markers: Absent (untrusted content is read directly into context)
  • Capability inventory: Bash (executing git, python, and asm tools), Write, and Edit (modifying files in the local repository)
  • Sanitization: Absent (content is used to inform prompts and edits without explicit filtering)
  • [COMMAND_EXECUTION]: The skill executes local CLI tools including 'asm', 'git', and 'python' to perform skill evaluation, version control synchronization, and mechanical validation. These operations are essential for its function as a development automation tool.
  • [EXTERNAL_DOWNLOADS]: The skill performs 'git fetch' and 'git pull' operations to synchronize the target repository with its remote origin. These network operations are directed by the user's repository configuration.
Audit Metadata
Risk Level
SAFE
Analyzed
May 16, 2026, 07:48 PM
Security Audit — agent-trust-hub — skill-auto-improver