The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local Node.js and PowerShell scripts to validate plans, render HTML, and generate PowerPoint skeletons. These operations are performed using standard system tools and COM automation for Microsoft PowerPoint. All commands are intended for local development workflows and do not attempt to elevate permissions or run arbitrary remote code.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes customer-supplied source materials, which presents a risk of indirect prompt injection.\n
Ingestion points: Customer-authored text, markup, and plans are read after passing through the scripts/preflight-sources.mjs utility.\n
Boundary markers: Instructions in SKILL.md require the agent to maintain a strict "Action boundary" when reporting preflight results, stating that source content never authorizes sensitive operations like network access or uploads.\n
Capability inventory: The skill can execute local scripts, automate PowerPoint via COM, and run a local HTTP server for previewing readouts.\n
Sanitization: The preflight-sources.mjs script performs regex-based scanning to detect and block malicious patterns such as instruction overrides, role spoofing, and credential requests. Furthermore, render-html.mjs implements HTML entity escaping for all generated content.\n- [SAFE]: No malicious obfuscation, persistence, or data exfiltration patterns were detected. The design emphasizes local execution, bindings for servers are restricted to 127.0.0.1, and the skill explicitly prevents the use of remote runtime conversion packages.

fde-readout