The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves reading and analyzing external manuscripts and scholarly literature stored in the library/markdown/ directory. These sources are considered untrusted data and could potentially contain instructions designed to influence the agent's behavior during persona construction or review generation.
Ingestion points: Manuscripts (Phase 0, 3) and literature files in library/markdown/ (Phases 1, 2, and 3).
Boundary markers: While the skill uses structured templates for output, it lacks explicit instructions to ignore or isolate embedded commands within the processed literature or manuscripts.
Capability inventory: The agent can execute shell commands via scripts, perform git operations for version tracking, and write files to the local project directory.
Sanitization: The skill does not implement specific sanitization or filtering for the content of ingested markdown files.- [COMMAND_EXECUTION]: The scripts/ingest.py script utilizes subprocess.run to execute docling and pandoc for document conversion. These calls are performed using list-style arguments and do not use a shell, but they operate on user-provided file paths.- [EXTERNAL_DOWNLOADS]: The scripts/ingest.py script fetches bibliographic metadata from established services (Crossref and Open Library) using identifiers extracted from documents. The scripts/rag.py script uses the sentence-transformers library, which may download pre-trained models from Hugging Face.

peer-reviewer