The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill interacts with live, third-party web content which represents a data ingestion surface for untrusted inputs. If a targeted website contains malicious text or instructions hidden in its HTML/markdown, it could manipulate the agent's subsequent browser actions.
Ingestion points: Web content fetched via the firecrawl scrape command and managed during firecrawl interact sessions inside SKILL.md.
Boundary markers: Absent. The instructions do not prescribe explicit boundary markers or system instructions to ignore embedded directives in the scraped text.
Capability inventory: The agent has access to executing firecrawl and npx firecrawl-cli commands in the environment via the shell.
Sanitization: Absent. No runtime input sanitization or filtering is described to clean web data before interaction.
[COMMAND_EXECUTION]: The skill utilizes platform configuration capabilities (allowed-tools) to execute local CLI commands (firecrawl * and npx firecrawl-cli *). While these commands belong to the tool's official vendor and represent expected functionality, executing shell commands based on external configurations or variables requires robust input validation at the execution environment layer.

firecrawl-interact