The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill is designed to process untrusted data from external sources, which could contain malicious instructions meant to manipulate the AI's behavior.
Ingestion points: External data enters the system through materialUrl (web pages), materialFile (uploaded documents like PDF, CSV, TXT), and materialText (user-pasted text) as specified in SKILL.md and references/form-schema.json.
Boundary markers: No explicit delimiters or 'ignore' instructions are present in the skill configuration to segregate external data from the agent's core instructions.
Capability inventory: The skill's runner script (scripts/run.py) has network capabilities, specifically making POST requests to the ai-skills.ai domain to execute tasks and poll for results.
Sanitization: There is no local validation, filtering, or sanitization of input data before it is transmitted to the remote API service.

firecrawl