The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/run_sample_calibration.py utilizes subprocess.check_call to create a Python virtual environment and install the requests library if it is missing from the environment. This is a standard self-healing dependency management pattern to ensure the script can run in restricted environments.
[DYNAMIC_EXECUTION]: To ensure the availability of dependencies, scripts/run_sample_calibration.py uses os.execv to replace the current process with a new instance of the script running within a temporary virtual environment. This is used exclusively for loading the requests library.
[INDIRECT_PROMPT_INJECTION]: The skill processes and uploads a bundled sample dataset (assets/sdg_08_2_sample_data_010926.zip). While this represents a theoretical attack surface, the risk is mitigated by explicit instructions.
Ingestion points: The agent extracts and reads synthetic warehouse data (videos, alignment JSON, and ground truth zip) from a local assets folder.
Boundary markers: The SKILL.md file contains explicit 'When to Use' and 'Do NOT use' instructions to prevent the agent from applying this skill to untrusted user data or RTSP streams.
Capability inventory: The skill set includes file system writes (unzip), HTTP requests to localhost via the requests library, and subprocess execution for environment setup.
Sanitization: The script performs validation on the input dataset, such as asserting a specific range of camera files (between 2 and 16) to prevent accidental processing of incorrect directories.
[COMMAND_EXECUTION]: The SKILL.md instructions include bash snippets for local port scanning (ports 8000-8009) to verify the microservice is ready and idempotent unzip commands for data extraction.

amc-run-sample-calibration