The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted content from multiple web-based sources (blogs, RSS feeds, GitHub releases), creating a surface for indirect prompt injection where malicious instructions could influence the agent's summary or briefing output.\n
Ingestion Points: Official vendor blogs, GitHub releases, RSS/Atom feeds, and user-provided public URLs.\n
Boundary Markers: While the skill uses structured parsing, it does not explicitly define boundary delimiters or specific "ignore embedded instructions" directives in its collection prompts.\n
Capability Inventory: Shell execution (Node.js), network collection, file system state management, and local browser rendering (Playwright).\n
Sanitization: Employs remark for markdown AST parsing and Zod for strict schema validation of the intermediate slides-data.js artifact, which helps mitigate simple injection-based layout or script corruption.\n- [COMMAND_EXECUTION]: The skill utilizes a Node.js-based pipeline to generate HTML, PDF, and PPTX presentations, executing local scripts with environment variables derived from user-selected profiles.\n
Evidence: AI_BRIEFING_PROFILE=\"$PROFILE\" node \"${CLAUDE_PLUGIN_DATA}/runtime/build/run.js\". The skill implements sanitization to ensure profile names are alphanumeric kebabs to prevent path traversal or injection.\n- [EXTERNAL_DOWNLOADS]: The setup process requires downloading and installing external Node.js packages and browser binaries (Chromium) to enable the report generation features.\n
Evidence: npm ci and npx playwright install chromium --only-shell.\n- [DATA_EXFILTRATION]: The skill performs automated outbound network requests to aggregate news items and validate the reachability of source links.\n
Details: Primarily targets official vendor sites, GitHub, and configured RSS feeds, but also supports user-supplied URLs.

generate