The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes paper abstracts retrieved from the arXiv API, which are untrusted external inputs. A maliciously crafted abstract could contain instructions intended to influence the subagent's extraction or the main agent's thematic synthesis. * Ingestion points: Paper abstracts are fetched via the arxiv_search.py script from the external export.arxiv.org API. * Boundary markers: The skill instructs agents to use structured prompts with markers like [Paper 1] and explicit field headers when delegating tasks to subagents. * Capability inventory: The skill has the ability to write files to the local filesystem (/mnt/user-data/outputs/), execute local scripts, use the task tool to spawn subagents, and call the present_files tool. * Sanitization: Subagent outputs are required to be in structured JSON format and are parsed by the main agent to maintain data integrity.
[EXTERNAL_DOWNLOADS]: The arxiv_search.py script performs network requests to the official arXiv API to retrieve metadata. This operation is essential for the skill's function and targets a well-known academic service.

systematic-literature-review