skill-installer

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill fetches data and software archives from GitHub infrastructure to perform its primary function.
  • [REMOTE_CODE_EXECUTION]: The skill installs external executable content (other skills) into the agent's local directory ($CODEX_HOME/skills). While it does not execute the downloaded content directly, it enables the introduction of unverified logic into the agent's runtime environment.
  • [COMMAND_EXECUTION]: The skill invokes the system's git command to perform repository cloning and sparse checkouts.
  • [INDIRECT_PROMPT_INJECTION]: The skill represents an attack surface for indirect prompt injection because it processes external GitHub URLs that could point to malicious repositories.
  • Ingestion points: Command-line arguments derived from user prompts in SKILL.md.
  • Boundary markers: Absent. No warnings are provided to verify the reputation of external repositories before installation.
  • Capability inventory: File system write access to skill directories and network access for downloads.
  • Sanitization: Includes basic path traversal protection (Zip Slip mitigation) in the extraction logic but does not audit the content of the downloaded skills.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 09:56 AM
Security Audit — agent-trust-hub — skill-installer