autobahn
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted user tasks that are explicitly categorized as 'risk-adjacent' (e.g., privacy, IP, security, stealth).
- Ingestion points: The skill reads user tasks and adjacent inputs during the 'FRAME' and 'CARVE' phases in SKILL.md.
- Boundary markers: It employs a 'scope-guard block'—a set of absolute exclusions and allowed alternatives—folded into the carved prompt to delimit the sub-agent's behavior.
- Capability inventory: The skill possesses the capability to spawn 'fresh, context-clean subagents' and write data to an 'archive destination' on the filesystem.
- Sanitization: The primary function of the skill is sanitization ('CARVE'), where it identifies risky items and replaces them with risk-free alternatives before passing a distilled prompt to the sub-agent.
- [DYNAMIC_EXECUTION]: The skill orchestrates the runtime creation and invocation of new agent contexts. In the 'RUN' phase, it spawns a 'fresh, context-clean subagent' (a separate task/process) and transfers a dynamically generated 'carved prompt' to it. This mechanism ensures that the execution environment is decoupled from the original session's reasoning and potentially risky inputs.
Audit Metadata