refining-workflows-from-artifacts
Pass
Audited by Gen Agent Trust Hub on Aug 7, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill defines a procedure for refining agent workflows based on external inputs such as execution results, reviews, and artifacts. This creates a surface for indirect prompt injection where malicious content within those artifacts could influence the agent to introduce unsafe or unauthorized changes into a workflow diff.
- Ingestion points: The skill explicitly instructs the agent to use 'Artifact produced' and 'Review, test result, simulation result, or human concern' as primary inputs in
SKILL.md. - Boundary markers: There are no instructions or delimiters defined to isolate untrusted external data from the agent's core logic or to warn the agent about potentially malicious instructions embedded in the artifacts.
- Capability inventory: The skill's primary output is a 'Workflow Diff', which modifies instructions for other agents. If the agent has the capability to apply these diffs to files, an attacker could manipulate the refinement process to inject malicious instructions into the target workflow.
- Sanitization: The procedure lacks any validation or sanitization steps for the ingested data, relying entirely on the agent's internal classification logic.
Audit Metadata