The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches paper titles and abstracts from external, user-contributed repositories (arXiv and bioRxiv) and processes them directly within the AI's context for evaluation and summarization. This creates an attack surface where a maliciously crafted paper abstract could attempt to influence the agent's behavior during the review process.
Ingestion points: Metadata is retrieved via scripts/arxiv_search.py from export.arxiv.org and api.biorxiv.org.
Boundary markers: The instructions in SKILL.md and references/writing-style.md do not specify the use of protective delimiters or "ignore embedded instructions" directives when handling the fetched content.
Capability inventory: The agent is authorized to execute a local Python script and perform file system writes to the daily paper/ directory.
Sanitization: The fetching script performs basic whitespace normalization but does not implement sanitization or filtering to remove potential prompt injection sequences from the external text.
[COMMAND_EXECUTION]: The skill requires the agent to execute a provided Python script (scripts/arxiv_search.py) with user-provided queries. The script performs network operations to fetch data from well-known scientific databases and writes results to a local file. This is a standard part of the skill's intended functionality.
[EXTERNAL_DOWNLOADS]: The scripts/arxiv_search.py script utilizes the feedparser library to process Atom feeds from arXiv. While feedparser is a standard, widely-used library for this purpose, it is an external dependency that the agent is expected to have available in its environment.

daily-paper-generator