guard-gauge-diagnostic
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is built to ingest and analyze untrusted data from external sources, creating a surface for embedded instructions that could attempt to influence agent behavior.
- Ingestion points: The workflow accepts "conversational dump, messy notes, CRM table, or uploaded PDF" as primary inputs (SKILL.md, Step 1) and uses patterns to extract signals from these unstructured sources (references/input-rules.md, Rule 3 and 4).
- Boundary markers: There are no explicit boundary markers or instructions to treat inputs as untrusted data, though the skill utilizes a "Question Budget" to limit user-agent loops.
- Capability inventory: The skill is primarily a reporting and diagnostic tool focused on text output. It does not contain evidence of and is not configured to use risky capabilities such as subprocess execution, file system modifications, or arbitrary network access.
- Sanitization: No explicit sanitization, character filtering, or data validation logic is documented; the agent is instructed to "map headers by meaning" and detect risk signals directly from natural language.
Audit Metadata