The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/skill-seekers-update.sh is designed to fetch a source code archive from an external, non-whitelisted GitHub repository (yusufkaraaslan/Skill_Seekers), extract it, and synchronize the files into the skill's local directory. This provides a mechanism for arbitrary remote code updates that could alter the tool's behavior after deployment.
[EXTERNAL_DOWNLOADS]: The skill performs automated network operations across several components: 1. scripts/skill-seekers-bootstrap.sh installs multiple Python dependencies from PyPI. 2. src/skill_seekers/cli/doc_scraper.py and github_scraper.py fetch documentation content and repository data from the internet. 3. scripts/skill-seekers-update.sh performs a direct download of a code archive from GitHub.
[COMMAND_EXECUTION]: The MCP server implementation in src/skill_seekers/mcp/server.py uses the subprocess module to construct and execute system commands for scraping and packaging skills. These commands are built using parameters passed through the agent interface, representing a standard but high-privilege capability.
[PROMPT_INJECTION]: As a generative tool for AI agent instructions, this skill exhibits a high vulnerability to indirect prompt injection (Category 8). According to the evidence chain: 1. Ingestion points: doc_scraper.py (web content), github_scraper.py (READMEs, issues), and pdf_scraper.py (PDF text). 2. Boundary markers: The generated SKILL.md files include source headers (see unified_skill_builder.py), but these are insufficient to prevent adversarial instructions in the source material from being parsed as agent rules. 3. Capability inventory: The tool can write to the local file system and execute subprocesses (mcp/server.py). 4. Sanitization: There is an absence of robust logic to sanitize natural language or executable instructions from scraped sources, allowing malicious content in third-party documentation to potentially manipulate agents that later use the generated skills.

skills-skills