The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to install dependencies and run code in the background.
Evidence: Instructions to run python3 -m pip install "msgpack>=1.1,<2" and python3 references/mock_adapter.py to host local adapters.
[DYNAMIC_EXECUTION]: The instructions guide the agent to perform on-the-fly code generation and execution.
Evidence: The skill explicitly directs the agent to create new Python scripts by modifying provided reference code (references/adapter_server.py, references/mock_adapter.py) and executing them at runtime.
[PERSISTENCE]: The agent is instructed to modify a local configuration file to register devices.
Evidence: Instructions to update or create ~/.qwen-mm-plugins/mhs-devices.json to persist adapter registrations across sessions.
[EXTERNAL_DOWNLOADS]: The skill provides links to external resources for extended functionality.
Evidence: References to a cookbook and documentation hosted on qwenlm.github.io, which is an official vendor domain.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from physical sensors and cameras via mhs_read which serves as a potential injection surface.
Ingestion points: The mhs_read tool retrieves image frames, sensor values, and text logs (documented in SKILL.md).
Boundary markers: The skill requires mhs_meta_info checks and defines safety limits, but does not specify sanitization for the content of the data read.
Capability inventory: The agent can actuate physical hardware via mhs_write, execute shell commands, and write to the filesystem.
Sanitization: No explicit sanitization or filtering of incoming sensor data strings or captions is described.

qwen-mm-plugins-mhs