research
Pass
Audited by Gen Agent Trust Hub on Jul 26, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection because it is designed to ingest untrusted data from the web (search engines, blogs, social posts) and use it to recommend modifications to the agent's core instructions, scripts, and configuration files.
- Ingestion points: Workflow Step 3 in
SKILL.mdand the discovery leads section inreferences/evidence-policy.mdspecify retrieving data from search results and other non-canonical external sources. - Boundary markers: The skill explicitly instructs the agent to treat discovery layers as non-authoritative and requires checking canonical sources for every decision-relevant claim.
- Capability inventory: In
references/workflow-integration.md, the skill describes modifying critical files such asAGENTS.md, reusable skills, plugins, and CI/CD hooks. - Sanitization: No programmatic sanitization is defined, but the workflow mandates validation of results, rollback conditions, and explicit labeling of weakly supported changes as experiments.
Audit Metadata