skillpack-admin

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted metadata from external source files during the skill discovery process.
  • Ingestion points: The discoverSkills function in scripts/lib.mjs reads the content of SKILL.md files from audited directories.
  • Boundary markers: None present; the script parses the raw content of the markdown files to extract YAML frontmatter.
  • Capability inventory: The skill has significant file system capabilities, including recursive directory creation and file copying through copyTreeOverlay in scripts/lib.mjs.
  • Sanitization: The skill applies a strict regex pattern (NAME_PATTERN) to validate extracted skill names before using them in file paths, which effectively prevents directory traversal or command injection via this metadata.
  • [COMMAND_EXECUTION]: The skill includes scripts that perform extensive file system operations (read, write, copy, hash) based on user-provided command-line arguments.
  • Evidence: The scripts/adopt.mjs and scripts/lib.mjs files implement logic to copy entire directory trees (copyTreeOverlay).
  • Mitigation: The implementation includes multiple security assertions, such as assertBoundedTarget to prevent operations on the filesystem root, and assertNoSymlinkAncestors / assertNoSymlinkBelow to prevent symbolic link attacks.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 01:49 PM