simulate-reviewers
Pass
Audited by Gen Agent Trust Hub on Jun 30, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local Python scripts (
scripts/review_form.pyandscripts/aggregate_scores.py) to generate review forms and calculate decision risk. These scripts use only Python standard libraries, perform deterministic logic, and do not make network calls. - [DATA_EXFILTRATION]: The skill explicitly includes a guardrail to process paper text transiently. It forbids copying paper content into the repository or storing it locally, quoting at most one sentence per finding to ensure the user's intellectual property remains protected.
- [EXTERNAL_DOWNLOADS]: While the process mentions fetching the live CFP URL, this is a manual instruction for the agent to verify facts against official conference websites (e.g., OpenReview, conference homepages). No automated downloads of executable code are present.
- [PROMPT_INJECTION]: The instructions contain guidelines to prevent the agent from being coerced into predicting real-world outcomes or fabricating data. It uses personas as analytical archetypes rather than a means to bypass safety filters.
Audit Metadata