coreweave-core-workflow-b
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from training logs which could theoretically contain malicious instructions designed to influence the agent.
- Ingestion points: In
SKILL.md, the commandkubectl exec -it $(kubectl get pod -l job-name=llm-finetune -o name) -- cat /checkpoints/training_log.json | tail -5reads content from a file residing on a persistent volume. - Boundary markers: There are no explicit delimiters or "ignore instructions" warnings for the agent when processing the content of these logs.
- Capability inventory: The skill allows for
Bash(kubectl:*)execution andWrite/Editoperations, which could be targeted by an injection attack. - Sanitization: The skill does not implement sanitization or validation of the log content before it is read into the agent's context.
Audit Metadata