muse-code-cli
Fail
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches an installation script for the Muse Code CLI from Meta's official developer domain (https://dev.meta.ai/install.sh).
- [REMOTE_CODE_EXECUTION]: Executes the downloaded installation script by piping it directly to the shell (sh).
- [COMMAND_EXECUTION]: The terminal command execution pattern allows the skill to run the Muse binary and associated scripts.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided task descriptions (briefs) that are passed to an external coding agent, creating a surface for indirect prompt injection. 1. Ingestion points: User-provided prompt files processed in SKILL.md. 2. Boundary markers: Instructions in SKILL.md recommend isolated git worktrees and specific configuration flags. 3. Capability inventory: The skill uses terminal, read_file, and write_file tools across all files. 4. Sanitization: Input validation for external content is absent.
- [DYNAMIC_EXECUTION]: The skill uses Python script templates to query a local SQLite database and parse JSONL logs to track session state as shown in references/sessions-and-jsonl.md.
Recommendations
- HIGH: Downloads and executes remote code from: https://dev.meta.ai/install.sh - DO NOT USE without thorough review
Audit Metadata