The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses docker ps and docker inspect commands to identify and verify the status of the GPU-enabled container where the pipeline runs. This is standard environment verification for a deployment tool.
[DYNAMIC_EXECUTION]: The skill orchestrates the transfer and execution of multiple Python scripts (e.g., collect_env_info.py, run_offline_inference.py) into Docker containers. This is the core functional mechanism of the pipeline to perform environment probes and model verification tests.
[INDIRECT_PROMPT_INJECTION]: The orchestrator's gate logic depends on JSON status reports produced by sub-scripts. This creates a surface where external data influences the pipeline's execution path, although the impact is restricted to the specific predefined states of the deployment orchestrator.
Ingestion points: Reads JSON output from scripts such as validate_packages.py, run_offline_inference.py, and diff_analysis.py via sub-skills.
Boundary markers: None explicitly defined in the orchestrator level; relies on the internal structure of the pipeline state schema.
Capability inventory: Capability to execute shell commands (docker) and run Python scripts inside containers across multiple steps.
Sanitization: None specified beyond parsing script output as JSON to update the pipeline state schema.

flagrelease-entrance-flagos