wait-what
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to process and utilize information from external files, specifically GLOSSARY.md and GLOSSARY-MAP.md, to modify its output. This creates a potential surface where adversarial instructions placed in those files could influence the agent's execution.
- Ingestion points: Instructions in SKILL.md direct the agent to read from GLOSSARY.md and follow GLOSSARY-MAP.md.
- Boundary markers: Absent. No delimiters or specific warnings are provided to help the agent distinguish between data and instructions within the glossary files.
- Capability inventory: The skill is focused on text re-pitching and includes a disable-model-invocation flag in its frontmatter, which restricts its ability to execute more complex actions.
- Sanitization: Absent. The skill does not describe any method for validating or filtering the content of the external files.
- [PROMPT_INJECTION]: The skill body and metadata use command-like language such as "Stop," "Wait," and "Re-pitch that" to override the current context and force a re-evaluation of the agent's previous response. While part of the intended functionality, these markers represent prompt injection patterns.
Audit Metadata