kinetograph-studio
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill starts a local editor environment using the
npx kinetograph studiocommand, which executes a package from the npm registry. It also supports an--agent <command>flag that allows specifying an arbitrary command to run as the chat agent. - [EXTERNAL_DOWNLOADS]: Execution of
npx kinetographinvolves downloading the package and its dependencies from the public npm registry at runtime. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the Kinetograph Studio UI, creating a surface for indirect prompt injection where user-provided content could influence agent behavior.
- Ingestion points: Data is read from
plan/*.jsonfiles (populated by user interaction with custom panels) and from 'Markup notes' messages which contain user-provided comments. - Boundary markers: No specific delimiters or instructions are used to separate user-provided content from the agent's instructions when reading the plan files or notes.
- Capability inventory: The skill can execute shell commands via
npx, write files to the local filesystem (inpresets/,plan/, andpublic/refs/), and modify studio state via MCP tools (studio_set_props). - Sanitization: There is no evidence of sanitization or validation of the content ingested from the Studio UI before the agent is instructed to "Fix the notes" or "build from [the plan]".
Audit Metadata