The Agent Skills Directory

[SAFE]: The skill uses environment variables (SEMANTIC_SCHOLAR_API_KEY) for authentication rather than hardcoded credentials, adhering to security best practices for secret management.
[EXTERNAL_DOWNLOADS]: The skill communicates with the official Semantic Scholar API (api.semanticscholar.org) and suggests the use of the Exa MCP service for paper discovery. These are well-known technical and academic services.
[PROMPT_INJECTION]: The bundled subagent (deep-paper-researcher.md) includes specific instructions to prioritize the user's verbatim request over potentially misinterpreted date windows or paraphrasing introduced by a calling agent, serving as a defensive measure in multi-agent workflows.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it processes external academic literature (abstracts and snippets).
Ingestion points: Data is fetched from the Semantic Scholar API and Exa MCP results.
Boundary markers: The subagent instructions do not explicitly define delimiters for external paper content.
Capability inventory: The agent can execute local Python scripts and perform file read/write operations to manage research results.
Sanitization: The skill distills and summarizes technical content; no explicit escaping of paper metadata is performed in the scripts.
[COMMAND_EXECUTION]: The skill executes locally provided Python scripts (ss_client.py, citation_graph.py) to perform structured research and graph traversal.
[DATA_EXFILTRATION]: No unauthorized data exfiltration was detected. The skill transmits only necessary search parameters to designated research APIs and follows 'token hygiene' practices by storing large results in local files.

semantic-scholar-deep