The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted manuscripts provided as text or PDF files, which are then stored in output/PAPER.md. This creates a surface for indirect prompt injection, where malicious instructions within the manuscript could influence downstream agents (such as claims-extractor) that read the generated file.
Ingestion points: scripts/run.py reads user-provided manuscript files from the workspace.
Boundary markers: The skill does not implement delimiters or specific "ignore instructions" markers when writing content to output/PAPER.md.
Capability inventory: The script scripts/run.py has file system write capabilities to the output directory.
Sanitization: No content sanitization or instruction filtering is performed on the ingested text during extraction or storage.

manuscript-ingest