sem-run
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied sketches and program definitions to generate semantic outputs.
- Ingestion points: Processes content from
program.mdand natural language inputs as defined inSKILL.md. - Boundary markers: The skill requires subagents to treat all program text as data and utilizes "no-history" workers to isolate the context.
- Capability inventory: Subagents are limited to generating semantic results and are explicitly prohibited from performing external effects or repository modifications.
- Sanitization: All outputs and intermediate artifacts are contained within a dedicated run directory with relative linking.
- [DYNAMIC_EXECUTION]: The skill dynamically generates prompt artifacts and orchestrates the execution of semantic applications.
- Runtime behavior: It interprets program sketches to create a plan, which is then executed by independent worker subagents.
- Mitigation: The framework enforces a strict "run contract" that isolates worker environments and limits their access to only authorized inputs.
Audit Metadata