The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes structured data from various external providers including APIs, licensed publishers, and government databases, which presents an attack surface for indirect prompt injection if those sources contain malicious instructions.
Ingestion points: Data is ingested through the firecrawl_scrape MCP tool and the firecrawl scrape CLI command.
Boundary markers: The instructions do not define specific delimiters for separating provider data from the agent's internal reasoning.
Capability inventory: The agent has the ability to execute Bash commands and call additional MCP search/scrape tools based on retrieved data.
Sanitization: No explicit sanitization or filtering of the structured records is described in the prompt instructions.
[EXTERNAL_DOWNLOADS]: The skill uses npx firecrawl *, which fetches the official Firecrawl package from the NPM registry to provide CLI functionality when the MCP is unavailable.
[COMMAND_EXECUTION]: The skill instructions allow the agent to execute shell commands via the firecrawl CLI for tasks such as searching, inspecting tool contracts, and scraping data.
[DATA_EXFILTRATION]: The CLI fallback documentation includes a feedback mechanism (firecrawl alexandria feedback) that transmits metadata, including targeted URLs, ratings, and task rationales, to the vendor's endpoint for service improvement.

firecrawl-alexandria