The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external data from meeting transcripts, emails, and call recordings provided by the user (see SKILL.md, Step 0.2 and Phase 1, Step 7). This creates a surface for indirect prompt injection where malicious instructions embedded in these materials could influence the agent's behavior. Ingestion points: Data enters the context via uploads in Step 0.2 and transcript analysis in Phase 1. Boundary markers: No delimiters or instructions to ignore embedded commands are specified. Capability inventory: The skill has the ability to generate and write executable code (HTML/JS) and perform searches in Phase 6. Sanitization: There is no mention of sanitizing or escaping the content from these materials before they are interpolated into the generated assets.
[COMMAND_EXECUTION]: In Phase 6, the skill generates self-contained HTML files that include JavaScript and CSS. This code is constructed using inputs from potentially untrusted external sources (the transcripts), creating a risk that malicious script fragments could be injected into the final output and executed when the asset is viewed in a browser.

create-an-asset