The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from Overleaf source files and external project websites. This content is ingested by the agent to generate the poster's HTML and to perform layout optimization via browser automation. A malicious actor could embed instructions in these sources to manipulate the agent's behavior.
Ingestion points: LaTeX source files in the overleaf/ directory, reference PDFs in the references/ directory, and external URLs provided for logo discovery.
Capability inventory: Writing local files to the poster/ directory, performing network requests to download institutional logos, and executing browser automation scripts via Playwright.
Boundary markers: No explicit delimiters or specialized instructions are used to distinguish between user-supplied data and agent instructions during processing.
Sanitization: No sanitization or validation of the processed LaTeX source or the external website content is described in the skill documentation.
[EXTERNAL_DOWNLOADS]: The skill setup involves cloning code from external repositories on GitHub and Overleaf. Additionally, the agent fetches institutional logos from user-specified project websites at runtime, and the generated poster application loads the React framework from a public CDN.
[COMMAND_EXECUTION]: The skill utilizes Playwright to automate layout verification and download logos. This requires the agent to execute shell commands and Node.js scripts to control a headless browser environment.
[DYNAMIC_EXECUTION]: The layout optimization process involves executing JavaScript within a browser context. The agent uses the posterAPI to programmatically interact with the generated HTML, facilitating dynamic card swapping and measurement of layout efficiency.

posterskill-academic-posters