The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs legitimate network operations to fulfill its primary purpose:
Fetches academic conference poster specifications (dimensions, orientation) from official venue websites.
Downloads the Chromium browser via Playwright (python3 -m playwright install chromium) to enable print-emulated PDF generation.
Downloads the MathJax tex-svg.js library for localized mathematical formula rendering.
[COMMAND_EXECUTION]: Executes various local utility scripts and system tools required for PDF processing and layout validation:
Invokes run_gates.py, render_preview.py, and poster_check.py from the skill's own directory.
Uses system tools such as pdftoppm, inkscape, or pdf2svg for figure extraction and conversion.
[DATA_EXPOSURE]: Accesses user-provided research papers (LaTeX or PDF) to extract text and figures for poster content. The skill explicitly warns against fabricating data and implements a provenance manifest for all extracted assets.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (academic papers) which presents an indirect injection surface.
Ingestion points: Research paper source files (.tex, PDF) read in Phase 1 and 2.
Boundary markers: None explicitly defined in the spawn_agent prompts; content is passed via file paths.
Capability inventory: The skill has access to Bash, Write, and Edit tools via the primary agent, and can spawn sub-agents for review.
Sanitization: No specific text sanitization is mentioned, but the use of "fresh" reviewer agents (Phase 1 and 6) mitigates cross-context injection risks by resetting session memory.

paper-poster-html