The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external material (proposals, specifications) to generate a decision instrument. Findings: 1. Ingestion points: Material submitted for a decision extracted in SKILL.md; 2. Boundary markers: Instructions to the agent to treat input as data and use mandatory citations; 3. Capability inventory: Local HTML file writing via scripts/render.mjs; 4. Sanitization: The render script implements a robust HTML entity escaping function (esc) applied to all extracted text.
[COMMAND_EXECUTION]: The skill instructs the agent to execute a provided local Node.js script (scripts/render.mjs) to generate the final output. This script is restricted to local filesystem operations, reading a specific JSON schema and writing an HTML file without network access.
[DYNAMIC_EXECUTION]: The rendering script generates an HTML file containing a serialized JavaScript function for interactive verdict calculation. The generated code is static and internal to the skill, and all data inserted into the HTML is properly escaped to prevent the execution of malicious payloads.

re0-decide