The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script (dag_validator.py) using uv run to perform graph analysis on task dependencies. It also utilizes latexmk to compile LaTeX files containing verbatim reviewer comments.
[EXTERNAL_DOWNLOADS]: The protocol includes instructions to install the networkx Python package from standard registries to support the dependency validation script.
[PROMPT_INJECTION]: The skill processes untrusted data from external PDF files containing referee reviews, creating an attack surface for indirect prompt injection.
Ingestion points: The reviews-original.pdf is read in Phase 2 and processed into atomic tasks.
Boundary markers: The skill lacks explicit boundary markers or instructions to treat the PDF content solely as data.
Capability inventory: The agent has capabilities for file system operations, script execution, and LaTeX compilation.
Sanitization: No sanitization or filtering is applied to the extracted reviewer text before it is interpolated into internal tracking and planning documents.

strategic-revision