The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data (such as code diffs, logs, issue bodies, and web content) to generate HTML artifacts. This creates an attack surface where malicious instructions embedded in the processed data could attempt to influence agent behavior. \n- Ingestion points: External data is processed in templates like the Review Workbench (diffs) and Evidence Dossier (research sources). \n- Boundary markers: The skill guidelines and security references mandate treating external content as untrusted and using structured formats to separate data from instructions. \n- Capability inventory: The agent has Write access to create artifacts and Bash access to execute a local Python validation script. \n- Sanitization: The skill explicitly instructs not to execute instruction-like text found in external content and to quote/summarize such content as data only. \n- [COMMAND_EXECUTION]: The skill workflow involves executing a local Python script (scripts/check_html_artifact.py) via the Bash tool to validate generated HTML artifacts. This script ensures artifacts are self-contained and accessible by detecting and blocking external network requests.

html-artifact