The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to pass the user's full request verbatim to shell commands (e.g., uv run --python 3.14 SKILL_DIR/scripts/clean_html.py PAGE.html). By explicitly directing the agent not to pre-parse or rephrase file paths or arguments, the skill increases the risk of command injection if the input contains shell metacharacters.\n- [DATA_EXFILTRATION]: The skill allows the agent to read from and write to arbitrary file paths provided in the user's request. This capability allows for the exposure of sensitive system files (e.g., configuration files, credentials) if the agent is not restricted to a specific directory sandbox.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from external HTML files and feeds the extracted data back into the agent's context. A malicious HTML file could contain instructions designed to override agent behavior or deceive the user.\n
Ingestion points: HTML files are processed by scripts/clean_html.py and scripts/extract_metadata.py.\n
Boundary markers: None. The skill does not instruct the agent to ignore instructions embedded within the extracted data.\n
Capability inventory: The skill uses uv run to execute Python scripts, which perform file read and write operations.\n
Sanitization: scripts/clean_html.py removes tags like <script> and <style>, but it does not sanitize the text content or metadata values for potential malicious instructions.

scrape-analyze-page