The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to gather research from external web sources via Firecrawl and synthesize that data into new AI agent skills (SKILL.md files). This creates an attack surface where malicious content on the public internet could attempt to inject instructions into the resulting skill artifacts.
Ingestion points: External research data is scraped and saved to the docs/research/skills/ directory before being processed by the 'write' phase.
Boundary markers: The skill does not currently implement explicit delimiters or 'ignore' instructions when interpolating gathered research into the synthesized skill content.
Capability inventory: The orchestrator has the capability to execute shell commands, run Python scripts, write files to the local filesystem, and invoke other slash commands in the agent's environment.
Sanitization: The format_skill_research.py script focuses on cleaning UI noise and formatting markdown but does not perform semantic validation or sanitization of the research content against adversarial instructions.
[EXTERNAL_DOWNLOADS]: The skill integrates with the Firecrawl service to perform web scraping and automated research. This is the intended behavior for the 'research' phase and allows the skill to fetch documentation and examples from the internet to inform skill creation.
[COMMAND_EXECUTION]: The orchestrator manages the skill lifecycle by executing local Python scripts (quick_validate.py, package_skill.py, format_skill_research.py) and various internal slash commands. This execution is scoped to the development workflow and uses standard Python libraries for file and YAML processing.

skill-factory