ableton-mcp
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides instructions for the user to clone and install an MCP server from an untrusted third-party GitHub repository (
github.com/jpoindexter/ableton-mcp). - [COMMAND_EXECUTION]: The skill instructs the agent to execute local system commands for operational tasks:
- Use of the
screencaptureutility to obtain visual feedback from the desktop (SKILL.md). - Execution of
gunzipandgrepto extract and search through the contents of Ableton project XML files (references/als-xml.md). - [INDIRECT_PROMPT_INJECTION]: The skill ingests and interprets user-controlled strings, creating a surface for potential indirect injection attacks through malicious project content.
- Ingestion points:
references/als-xml.md(Reads arrangement content and MIDI note events from.alsproject files);SKILL.md(Reads clip labels to infer harmony and project structure). - Boundary markers: Absent; the agent is instructed to treat clip names and project XML as current truth without delimiters or warnings regarding embedded instructions.
- Capability inventory: MIDI note manipulation (
add_notes_to_clip), track/scene creation via TCP socket commands, and local shell execution (gunzip,screencapture). - Sanitization: Absent; no escaping or validation of ingested project metadata is described before interpolation into the agent's reasoning process.
Audit Metadata