devfleet
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides an orchestration layer that ingests prompts and dispatches them to sub-agents. This creates a surface for indirect prompt injection where instructions embedded in project plans or mission descriptions could influence sub-agent behavior.
- Ingestion points: The
plan_project(prompt)andcreate_mission(..., prompt=...)tools ingest untrusted natural language data. - Boundary markers: Absent. The documentation does not specify the use of delimiters (e.g., XML tags or triple backticks) to separate instructions from untrusted content.
- Capability inventory: Dispatched agents operate in git worktrees and have access to full tooling for coding, testing, and merging changes.
- Sanitization: Absent. There is no evidence of input validation or escaping for the prompts passed to missions.
- [EXTERNAL_DOWNLOADS]: The setup instructions direct users to download and install a server from a repository (
https://github.com/LEC-AI/claude-devfleet) that is not identified as a trusted source. - [COMMAND_EXECUTION]: The skill facilitates parallelized command execution via multiple agents. While this is the primary purpose of the skill, it involves automated git operations (worktree creation, merging) and sub-process execution by the orchestrated agents.
Audit Metadata