The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes system utilities such as curl, wget, and npx playwright screenshot to fetch HTML and capture visual evidence from target designs. It also executes a bundled Python script scripts/score.py to perform deterministic scoring based on detected design flaws.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it is designed to ingest and process untrusted content from external websites and codebases.
Ingestion points: The target parameter in SKILL.md allows users to provide live URLs or code paths which are then crawled and analyzed by the agent.
Boundary markers: None identified; the instructions lack explicit delimiters to separate the evaluation instructions from text content found on the target websites.
Capability inventory: Network access (browser automation), file system writes (generating reports), and local script execution.
Sanitization: The skill does not describe specific sanitization or filtering for the textual content of the audited designs, which could potentially contain malicious prompts.

slop-eval