analyze-data
Pass
Audited by Gen Agent Trust Hub on May 18, 2026
Risk Level: SAFE
Full Analysis
- [DATA_EXFILTRATION]: The skill transmits local datasets to a remote endpoint ('asta-gateway') for analysis. This behavior is documented as the core functionality and is handled via the 'allenai' vendor's own CLI tools and infrastructure.
- [COMMAND_EXECUTION]: The skill utilizes the 'asta' CLI and 'jq' to submit analysis tasks, poll for completion, and index results. It also uses the 'open' command to display generated artifacts to the user. These commands are consistent with the skill's purpose and the author's verified toolset.
- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests and processes untrusted user-provided datasets (e.g., CSV files).
- Ingestion points: User-specified datasets and natural language analytical requests.
- Boundary markers: Present. The agent is instructed to synthesize a 'tightened analytical question' and present it for explicit user approval before submitting the task.
- Capability inventory: The skill uses the 'asta' CLI for network submission and local artifact management, and 'open' for results display.
- Sanitization: The workflow relies on mandatory human-in-the-loop confirmation ('Step 2') to validate the analysis plan before execution.
Audit Metadata