The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface because it is designed to ingest and interpret data from untrusted sources, such as public website HTML, local source code, and analytics exports.
Ingestion points: Public URLs, local HTML/source files, and analytics exports as specified in the 'Operating Modes' in SKILL.md.
Boundary markers: None identified; the skill does not define specific delimiters or instructions to ignore embedded commands within the analyzed site data.
Capability inventory: The skill instructions indicate capabilities for file system access (reading local project paths) and network access (crawling public URLs).
Sanitization: No sanitization or escaping mechanisms are described for the data being analyzed, although safety rules provide guidelines on what should not be stored.
[COMMAND_EXECUTION]: SKILL.md contains validation commands that execute Python scripts located in the $CODEX_HOME directory. These are intended for development-time linting and validation of the skill's own structure.
[DATA_EXFILTRATION]: The skill incorporates explicit safety protocols to prevent the accidental exposure of sensitive information. It instructs the agent not to store private tokens, cookies, or .env values in the generated audit reports.
[EXTERNAL_DOWNLOADS]: The skill is configured to perform network operations for crawling public URLs; however, these actions are governed by a required approval-gated workflow to ensure user oversight.

existing-site-analyzer