auto-skill

Warn

Audited by Gen Agent Trust Hub on Aug 4, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The script scripts/skill-seekers-update.sh is designed to download a compressed archive from a third-party GitHub repository (https://codeload.github.com/yusufkaraaslan/Skill_Seekers/tar.gz/main). This repository is not associated with a known trusted vendor or organization.
  • [REMOTE_CODE_EXECUTION]: The script scripts/skill-seekers.sh executes the code downloaded by the update script. It sets the PYTHONPATH to the downloaded source directory and runs the module using python3 -m skill_seekers.cli.main, effectively executing unverified external code within the agent's environment.
  • [COMMAND_EXECUTION]: The skill provides several Bash scripts (create-skill.sh, skill-seekers-bootstrap.sh, skill-seekers-import.sh, validate-skill.sh) that perform significant filesystem operations, including creating directories, using rsync to delete/overwrite files, and managing Python virtual environments with pip install.
  • [PROMPT_INJECTION]: As a meta-skill designed to ingest external documentation and GitHub repositories to generate new instructions, it is inherently vulnerable to indirect prompt injection. Malicious instructions embedded in the source material (e.g., a targeted website or repository) could influence the content of the generated SKILL.md file, potentially poisoning downstream skills.
  • Ingestion points: scripts/skill-seekers.sh reads external URLs, GitHub repositories, and PDFs.
  • Boundary markers: The generated templates in assets/ include boundary sections, but there are no markers or sanitization logic applied to the input data being scraped.
  • Capability inventory: The skill can execute shell commands, perform network requests (via curl), and write to the filesystem.
  • Sanitization: None detected. The validation script checks for structure and length but does not inspect the content for malicious prompt instructions.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 4, 2026, 02:05 PM
Security Audit — agent-trust-hub — auto-skill