The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): The skill requires installing firecrawl-cli@1.4.1 from npm. This is an external dependency from a non-whitelisted source.
DYNAMIC_EXECUTION (MEDIUM): The skill utilizes firecrawl browser execute with --python, --node, and --bash flags. While documented as running in a remote sandboxed environment, this provides a mechanism for script generation and execution based on agent-generated or user-provided input.
PROMPT_INJECTION (LOW): The skill is designed to ingest untrusted third-party data from the web (Category 8). This creates a surface for indirect prompt injection where malicious instructions in scraped content could influence the agent.
Ingestion points: firecrawl scrape, firecrawl search, and firecrawl crawl save data to the .firecrawl/ directory.
Boundary markers: The skill documentation advises the agent to use -o to write results to files and use incremental reading (grep/head) to avoid direct context injection.
Capability inventory: The agent can execute shell commands, read/write local files in the .firecrawl/ directory, and perform network operations via the CLI.
Sanitization: No programmatic sanitization is defined; the skill relies on instructional guidance for the agent to treat content as untrusted.
COMMAND_EXECUTION (LOW): The skill operates by executing shell commands. Improper sanitization of user-provided URLs or search queries could lead to local command injection if the agent does not properly quote arguments as instructed.

firecrawl