The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes and displays runbook logs which could contain malicious or misleading instructions.\n
Ingestion points: The view.py script reads state.json and progress.md from a user-specified run directory.\n
Boundary markers: SKILL.md instructs the agent to present output in code blocks without interpreting it, reducing the likelihood of the agent following instructions found within logs.\n
Capability inventory: The skill starts a local web server to serve log data and uses the marked library for markdown rendering.\n
Sanitization: The tool implements thorough security measures: view.py performs path normalization and validation to block directory traversal; the server enforces a strict CSP and validates the Host header to prevent cross-site attacks and DNS rebinding; and the markdown renderer is configured to escape HTML tags to prevent XSS.

runbook-viewer