The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from remote Overleaf projects, creating an attack surface where malicious LaTeX files could influence the agent's behavior.
Ingestion points: src/client.ts via downloadProject, src/mcp.ts via pull_project tool.
Boundary markers: Absent (processes raw LaTeX and text content).
Capability inventory: The skill has extensive capabilities including filesystem write/read and network operations to Overleaf domains.
Sanitization: While the CLI implementation in src/cli.ts correctly validates archive entry paths using resolveWithin to prevent path traversal (zip-slip), the MCP implementation in src/mcp.ts for the pull_project tool lacks this validation and directly concatenates entry names with the output directory.
[EXTERNAL_DOWNLOADS]: Communicates with overleaf.com or user-defined self-hosted instances to synchronize files, download PDFs, and fetch compilation artifacts.
[COMMAND_EXECUTION]: Executes local git commands through src/remote-helper.ts to implement native git remote functionality for Overleaf projects.

overleaf