The Agent Skills Directory

[SAFE]: The skill is a hardened implementation of a technical report generator that explicitly disables unsafe legacy features such as network requests, remote updates, and automatic browser interaction. It relies solely on built-in Node.js modules.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied data to generate HTML visualizations. It mitigates this risk through a comprehensive multi-layered defense: (1) Ingestion points: Data is received via stdin; (2) Boundary markers: Instructions in SKILL.md explicitly state that input data must not override agent authorization; (3) Capability inventory: Operations are limited to reading/writing within the .am-pages directory; (4) Sanitization: A robust HTML escaping function (esc()) is applied to all input, and the output includes a strict Content Security Policy (default-src 'none').
[COMMAND_EXECUTION]: The skill provides the agent with shell commands to invoke a local script (am.mjs). This script is audited to perform only safe file operations within a restricted workspace, avoiding shell injection and arbitrary execution.
[DATA_EXFILTRATION]: File access is restricted to a managed workspace (.am-pages) using path.resolve and fs.realpathSync to prevent path traversal. The script also uses O_NOFOLLOW and O_EXCL flags to prevent symlink attacks and accidental file overwrites.

answer-me-with-html