The Agent Skills Directory

[COMMAND_EXECUTION]: The skill employs multiple Python scripts (finalize_output.py, cleanup_intermediate.py, list_proposal_files.py) to automate its review lifecycle. These scripts execute file system tasks such as directory cleanup and file reorganization. All scripts include path validation checks to strictly enforce that file operations do not traverse outside the intended project root.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it processes untrusted user LaTeX files. 1. Ingestion points: Files matching scientific proposal patterns like *.tex within the proposal_path. 2. Boundary markers: Content is summarized and interpolated into expert prompts without explicit isolation delimiters. 3. Capability inventory: Execution of local Python helper scripts for file management and task orchestration via the parallel-vibe utility. 4. Sanitization: No validation or escaping is performed on proposal content before processing.
[SAFE]: The skill demonstrates secure implementation patterns, including the use of safe YAML loading for configuration and keeping sensitive data localized within the user's project structure with no unauthorized network transmissions.

nsfc-reviewers