The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates tasks by executing commands via subprocess.run to invoke other skills such as scillm and create-paper.
[PROMPT_INJECTION]: The skill documentation establishes a 'Pen Name Enforcement' policy that mandates specific identity mappings, acting as a behavioral constraint that overrides standard agent logic. Additionally, the skill is susceptible to indirect prompt injection from untrusted external content.
Ingestion points: Research paper text extracted from ArXiv in src/curriculum.py.
Boundary markers: Lacks explicit delimiters or instructions for the LLM to ignore embedded commands in paper content.
Capability inventory: Includes sub-process execution via uv run and file system writes for review reports and snapshots.
Sanitization: Relies on basic regex-based stripping of HTML and LaTeX tags.
[EXTERNAL_DOWNLOADS]: The skill retrieves paper content and metadata from well-known academic services including ar5iv.labs.arxiv.org and api.semanticscholar.org.

create-peer-review