The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes local shell and Python scripts to perform audit tasks. Evidence includes the execution of claude-seo run unlighthouse_run.py and claude-seo run lcp_subparts.py, as well as a reference to a local installation script extensions/unlighthouse/install.sh.
[EXTERNAL_DOWNLOADS]: The skill relies on external software components, specifically the unlighthouse npm package mentioned in the compatibility section of the SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The core functionality of auditing external websites introduces an attack surface for indirect prompt injection.
Ingestion points: The tool fetches and parses content from user-specified external URLs using the Unlighthouse CLI.
Boundary markers: There are no explicit delimiters or instructions defined in the skill manifest to treat the tool's output as untrusted content.
Capability inventory: The skill uses subprocess execution and file system reads (ci-result.json) to retrieve audit data.
Sanitization: The manifest mentions a url_safety pre-flight check performed by the unlighthouse_run.py wrapper script.

seo-unlighthouse