The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted external data via commands like @summarize (transcripts) and @coach (interview questions). This creates a surface for indirect prompt injection, where malicious instructions could be embedded in the input text to manipulate the agent's behavior or output. While the skill includes internal rules to filter logical noise such as compliments and hypotheticals, it lacks dedicated security sanitization to prevent the execution of injected instructions.
[DATA_EXFILTRATION]: The skill's CLI implementation sends user-supplied ideas and persona information to external AI providers, specifically OpenAI and Anthropic. This network activity is fundamental to the skill's purpose and targets well-known services, but it represents data egress to external platforms.
[EXTERNAL_DOWNLOADS]: The skill's package configuration includes dependencies on several well-known Node.js packages from the NPM registry, including promptfoo, markdownlint-cli, and better-sqlite3. These tools are standard for development, testing, and documentation linting, and originate from established sources.

personatwin