The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses docker run and docker exec to manage workflow lifecycles. These operations are restricted to the allocated Data Services (DS) container and are necessary for the skill's primary function of model training and data mining as described in the lifecycle instructions.
[DYNAMIC_EXECUTION]: The workflow allows for user-defined scoring and evaluation logic via implementation_files. The skill includes technical safeguards such as SHA-256 integrity hashing and 'lock' files to ensure that only approved code is executed within the pipeline, mitigating risks associated with dynamic loading.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data manifests and configuration files, creating a potential surface for indirect injection.
Ingestion points: target_manifest, source_payload_contract, and custom implementation_files specified in the workflow configuration.
Boundary markers: The system enforces immutability via data.lock.json and uses _SUCCESS markers to verify the integrity of stage outputs.
Capability inventory: File system write access, Docker container management, and Python subprocess execution.
Sanitization: The controller performs validation of implementation digests and input contracts before proceeding with mining or training stages.

tao-run-dinov3-ssl-deft