The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md and the implementation in capture_screenshot.py frequently use subprocess.run and direct shell commands like curl, ls, and npx to perform environment checks and execute browser automation tools.
[REMOTE_CODE_EXECUTION]: In scripts/capture_screenshot.py, the skill dynamically generates temporary JavaScript or Python scripts using string templates and executes them via subprocess. The provided URL and output paths are interpolated directly into these templates with minimal escaping, which could lead to code injection within the temporary script context if a maliciously crafted URL is processed.
[EXTERNAL_DOWNLOADS]: The skill requires and encourages the installation of external dependencies, specifically the playwright library and its associated browser binaries, through Node.js (npm) or Python (pip) package managers.
[PROMPT_INJECTION]: The skill facilitates indirect prompt injection (Category 8) by browsing and capturing content from arbitrary URLs provided by the agent or user.
Ingestion points: The url argument in scripts/capture_screenshot.py allows the skill to ingest data from external web sources.
Boundary markers: Not present; the skill treats captured content as binary evidence but lacks safeguards if an agent subsequently parses the content of the captured pages.
Capability inventory: The skill possesses significant local capabilities, including file system access and the ability to execute arbitrary subprocesses through its utility scripts.
Sanitization: There is no significant sanitization or validation of the input URL before it is used in code generation or browser navigation.

testing-evidence-collector