The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection due to the ingestion of untrusted data from the web.
Ingestion points: Untrusted data such as paper abstracts, titles, and web content is retrieved via WebFetch and WebSearch (invoked in SKILL.md) and processed in the agent's context.
Boundary markers: The skill does not implement boundary markers or instructions to ignore potential commands embedded in the fetched research papers or web snippets.
Capability inventory: The agent has access to powerful tools listed in SKILL.md, including Bash, Edit, Write, and the ability to call other agents, which could be exploited if malicious instructions are encountered in the external data.
Sanitization: There is no logic provided to sanitize or filter the content retrieved from external sources before it is recorded in the local workspace files like memory-bank.md or mind-graph.md.
[EXTERNAL_DOWNLOADS]: The skill performs necessary network requests to fetch research data.
Sources: It connects to the Semantic Scholar API (api.semanticscholar.org) and performs web searches to retrieve metadata for recent research papers.
Usage: The retrieved data is used to populate local markdown files within the wiki/queries/ directory.

lit-search