jinko-task-paper2model
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external scientific literature, supplements, and OCR data to populate the reproduction specification. This untrusted data is then used to influence model implementation and platform mutation.
- Ingestion points: Data is extracted from external PDFs, supplements, and OCR packets into the
reproduction-spec.jsonfile. - Boundary markers: The skill employs a strict JSON schema (
reproduction-spec.schema.json) and specific Markdown templates for reporting, which act as structural boundaries. - Capability inventory: The skill utilizes tools like
jinko-model,jinko-protocol,jinko-vpop, andjinko-trialto create and mutate artifacts on the Jinkō platform. - Sanitization: All specification data is validated by a local script (
paper2model_report.py) before platform mutation. The instructions explicitly require human visual review of OCR-derived content and equations to ensure fidelity. - [COMMAND_EXECUTION]: The skill instructs the agent to execute a local Python script (
scripts/paper2model_report.py) for validating specifications and rendering reports. - Evidence: The workflow includes commands like
python scripts/paper2model_report.py validate --spec reproduction-spec.jsonandpython scripts/paper2model_report.py render. - Context: This script is part of the skill's distribution and uses standard libraries to perform deterministic data processing and validation tasks.
Audit Metadata