The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data from academic papers which presents a potential vector for indirect prompt injection.
Ingestion points: Research paper metadata (titles, abstracts) retrieved from APIs (PubMed, arXiv, Semantic Scholar, etc.) and full-text content extracted from downloaded PDFs in the papers/ directory.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when the agent extracts methodology or summarizes paper content.
Capability inventory: The skill executes local Python scripts (scripts/initialize.py, scripts/bibtex.py), writes various files to the local filesystem within its workspace, and can interact with the gh (GitHub) CLI tool.
Sanitization: While filename sanitization is implemented in scripts/utils.py, there is no evidence of sanitization or filtering of the text content extracted from external papers before it is processed by the agent.
[COMMAND_EXECUTION]: The skill executes several local Python scripts to manage its operations:
scripts/initialize.py: Used to create the .make-research-plan workspace. This script includes a security check to ensure the target directory is a subdirectory of the current working directory, preventing directory traversal attacks.
scripts/bibtex.py: Generates BibTeX files from JSON metadata.
scripts/validate.py: Validates JSON outputs against predefined schemas.
The skill also mentions using the gh (GitHub) CLI tool for bug reporting to the author's repository (huangwb8/bensz-bugs) upon explicit user request.
[EXTERNAL_DOWNLOADS]: The skill is designed to fetch research paper metadata and PDF files from well-known and reputable scientific sources including PubMed, arXiv, IEEE Xplore, and Semantic Scholar.

research-plan