The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on a launcher script (scripts/litrun.py) that extensively uses subprocess.run to execute system commands. This is used for checking the toolchain, managing environment variables, and launching various external utilities.
[EXTERNAL_DOWNLOADS]: The skill is designed to download and install a large number of third-party Python packages and clone external GitHub repositories (such as GPT Researcher and STORM) during its operation.
[DYNAMIC_EXECUTION]: The launcher script assembles and executes shell commands and Python snippets at runtime based on definitions provided in recipes/recipes.json. This includes running code snippets via python -c and launching web servers (uvicorn, streamlit) for cloned projects.
[CREDENTIALS_UNSAFE]: The skill manages multiple sensitive API keys (e.g., OpenAI, Anthropic, Tavily, You.com) by storing them in a local .env file at ~/.lit-review-tools/.env. It also has functionality to propagate these credentials into the environment of cloned third-party tools when launching their user interfaces.
[REMOTE_CODE_EXECUTION]: Through its ui and run commands, the skill clones and executes code from remote repositories. While the catalog targets well-known research projects, this mechanism allows the execution of arbitrary logic from external sources at runtime.
[DATA_EXPOSURE]: The skill ingests and processes data from external scholarly APIs (arXiv, OpenAlex, PubMed) and local PDF files. This data is fed into LLMs for synthesis and Q&A, creating an attack surface for indirect prompt injection where adversarial content in a paper could attempt to influence the agent's behavior.
Ingestion points: scripts/fetch_arxiv.py, scripts/fetch_openalex.py, scripts/fetch_pubmed.py, and user-provided PDF paths.
Boundary markers: None explicitly implemented in the data retrieval scripts to delimit untrusted content.
Capability inventory: Full shell execution (via litrun.py), file system read/write, and network access.
Sanitization: No explicit sanitization or filtering of the content retrieved from external academic sources before processing.

literature-review-tools