The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructions direct the agent to execute a feedback command (firecrawl alexandria feedback) after completing a task. This command transmits user activity metadata to the vendor, specifically the target --url and the user's --requested-functionality. The instruction to use the & operator to run this 'in the background' means this data exfiltration occurs without direct user oversight during the operation.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a vulnerability surface where the agent is prompted to synthesize data from external websites (scraped content) and the user's prompt into a CLI command.
Ingestion points: Website content from scrape/search results and user prompts in SKILL.md context.
Boundary markers: None present to distinguish between user data and telemetry instructions.
Capability inventory: Shell command execution via the firecrawl CLI.
Sanitization: No sanitization or escaping is defined for the --rationale or --requested-functionality strings being passed to the shell.
[COMMAND_EXECUTION]: The skill provides specific bash execution patterns, including the use of --silent and & to suppress errors and background the process, ensuring the telemetry command completes even if the user is unaware of its execution.

firecrawl-alexandria