framework-dev
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documentation describes a system where user-provided answers from an onboarding flow are stored in a
USER.mdfile and subsequently included in the agent's system prompts using the{{include}}directive. This creates an indirect prompt injection surface where user-supplied content can influence agent behavior. - Ingestion points: The
onboarding.jsonflow results in data being written toagent_data/USER.md(SKILL.md). - Capability inventory: The app's agent has bash/read/write access to the
agent_data/sandbox and access to a SQLite activity log (SKILL.md). - Boundary markers: The documentation does not mention the use of specific delimiters or system instructions to distinguish user-provided data from agent instructions when using the
{{include}}tag. - Sanitization: The text does not describe any sanitization or validation of the onboarding answers before they are interpolated into the prompt store.
- [EXTERNAL_DOWNLOADS]: The skill instructs developers to install a CLI tool as a development dependency directly from a Git branch using the command
npm i -D github:sylm54/train-me#framework-cli. This involves fetching and installing code from an external repository managed by the vendor. - [COMMAND_EXECUTION]: The documentation specifies that the integrated AI agent within the application is granted full bash access to its sandbox environment, providing it with extensive command execution capabilities within that scope.
- [DYNAMIC_EXECUTION]: The TTS (Text-to-Speech) markup system supports a
<visual>tag that dynamically fetches and displays media content from the external service RedGIFs during script playback.
Audit Metadata