The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest data from external websites, creating a potential surface for indirect prompt injection via malicious cookie values, page snapshots, or console logs. 1. Ingestion points: evaluate_script for cookie retrieval, take_snapshot for DOM inspection, and list_console_messages in SKILL.md. 2. Boundary markers: The instructions do not specify the use of delimiters or safety warnings for processing untrusted web content. 3. Capability inventory: Includes evaluate_script, navigate_page, and click (SKILL.md). 4. Sanitization: No sanitization or validation of external content is performed.
[COMMAND_EXECUTION]: The skill provides a shell command for the agent to parse JSON audit reports locally. Evidence: Use of node -e to process /tmp/lh-report/report.json in SKILL.md.
[DYNAMIC_EXECUTION]: The skill utilizes browser-side script evaluation to interact with cookies and the document object model. Evidence: Frequent use of evaluate_script with cookieStore and document.cookie APIs in SKILL.md.

cookie-debugging