maa-cli-operate

Fail

Audited by Gen Agent Trust Hub on Nov 22, 2026

Risk Level: HIGHCOMMAND_EXECUTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill documentation explicitly supports arbitrary command execution through the action shell feature.
  • Evidence: SKILL.md states, "Treat action shell as arbitrary command execution on the selected controller."
  • [DYNAMIC_EXECUTION]: The skill facilitates the dynamic loading and execution of local Python code through the custom load command.
  • Evidence: references/command-contract.md includes the command maafw-cli --json custom load ./agent/main.py and notes that it "imports and executes Python code."
  • [EXTERNAL_DOWNLOADS]: The skill instructions utilize uvx to fetch and execute specific versions of the maafw-cli package from an external registry.
  • Evidence: references/command-contract.md defines the verified release as uvx --from maafw-cli==0.1.6 maafw-cli.
  • [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its broad capabilities and ingestion of untrusted data.
  • Ingestion points: The agent processes data from ocr, reco (recognition), and pipeline run commands, which ingest content from external screens and resource files.
  • Boundary markers: No specific delimiters or boundary enforcement instructions are provided to separate instructions from data within OCR or pipeline content.
  • Capability inventory: The skill possesses high-risk capabilities including action shell (arbitrary shell commands) and custom load (Python execution).
  • Sanitization: The skill relies on manual human inspection ("Inspect Python files before custom load") rather than automated sanitization or sandboxing.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Nov 22, 2026, 01:28 AM