skill-link

Warn

Audited by Gen Agent Trust Hub on Mar 7, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructions utilize shell-level logic such as 'ln -sfn' to create and force symlinks, and 'realpath' to resolve absolute paths. These operations directly modify the filesystem and the agent's environment configuration.
  • [DATA_EXFILTRATION]: The skill explicitly targets sensitive application directories located at '$HOME/.claude' and '$HOME/.codex'. Modifying or linking into these paths can expose tool-specific configurations or enable the persistence of unauthorized extensions.
  • [PROMPT_INJECTION]: The skill creates an indirect prompt injection surface by scanning external directories and reading 'SKILL.md' files during the installation and health-check workflows.
  • Ingestion points: User-specified source directories and contents of 'SKILL.md' files discovered during scanning.
  • Boundary markers: No delimiters or instructions are implemented to ignore embedded malicious content within the scanned files.
  • Capability inventory: File system linking, path resolution, and directory scanning.
  • Sanitization: Mentions path quoting to handle spaces but lacks validation for the integrity or safety of the content within the linked files.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 7, 2026, 04:21 PM