The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/skill-seekers-update.sh is designed to download a source code archive from a non-trusted third-party GitHub repository (yusufkaraaslan/Skill_Seekers) via curl and replace the contents of a local directory. This directory contains code that is subsequently executed by the agent using scripts/skill-seekers.sh.
[EXTERNAL_DOWNLOADS]: The skill includes scripts/skill-seekers-bootstrap.sh which executes pip install to fetch Python dependencies from external registries, and scripts/skill-seekers-update.sh which fetches remote archives from codeload.github.com.
[COMMAND_EXECUTION]: The skill provides several shell scripts (create-skill.sh, validate-skill.sh, skill-seekers-import.sh, skill-seekers.sh) that execute various system commands (sed, rsync, python3, mkdir, tar) and perform file system modifications based on user-provided inputs.
[INDIRECT_PROMPT_INJECTION]: The skill is specifically designed to extract and process domain material from external sources like documentation websites, GitHub repositories, and PDFs to create new instructions for AI agents. This ingestion workflow is vulnerable to indirect prompt injection if the source materials contain adversarial content intended to manipulate the agent's behavior during the generation process.
Ingestion points: Documentation sites, GitHub repositories, and PDF files as defined in SKILL.md.
Boundary markers: The SKILL.md file contains a Not For / Boundaries section providing some protection by defining what the skill will not do and requiring clarification for missing data.
Capability inventory: The skill has the ability to write files, run Python code, and access the network through its suite of utility scripts.
Sanitization: There is no evidence of explicit sanitization or instruction filtering for the content retrieved from external domain materials before it is used to generate new skill content.

auto-skill