research-pipeline-runner

Pass

Audited by Gen Agent Trust Hub on Mar 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes repository-local Python scripts (scripts/pipeline.py) to perform operations like workspace initialization, pipeline state management, and checkpoint verification. These scripts are used as deterministic helpers for the skill's coordination logic.
  • [EXTERNAL_DOWNLOADS]: The pipeline runner indicates network dependency for fetching papers from arXiv and verifying citations. These operations target well-known academic services and are directly related to the primary research functionality of the skill.
  • [PROMPT_INJECTION]: The skill processes untrusted data from user goals and external research documents, creating an indirect prompt injection surface.
  • Ingestion points: User-provided research topics/requests and content retrieved from external PDFs or arXiv searches.
  • Boundary markers: Uses a structured UNITS.csv file as an execution contract and implements mandatory human approval checkpoints (CHECKPOINTS.md) before progressing to prose generation.
  • Capability inventory: Subprocess execution for pipeline management and file-system write operations restricted to the workspaces/ subdirectory.
  • Sanitization: The skill relies on structured task breakdown and human review rather than explicit string sanitization to manage untrusted content.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 15, 2026, 01:59 PM