The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The file references/docker.md provides instructions for using sudo chown -R 10000:10000 and chmod -R 777 to resolve permission issues on host directory mounts for Docker containers. Additionally, it recommends mounting the host's /etc/passwd and /etc/group files into the container, which is a sensitive configuration that can lead to information disclosure.\n- [EXTERNAL_DOWNLOADS]: The references/datasets.md file documents that the preparation script for the UC2 dataset (prepare_dataset_uc2.py) automatically downloads content from an unverified third-party GitHub repository (abin24/Magnetic-tile-defect-datasets).\n- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests untrusted data from user-provided datasets and JSONL defect specifications.\n
Ingestion points: dataset_dir, defect_spec.jsonl, and clean_dir files.\n
Boundary markers: None detected in the instructions for processing these files.\n
Capability inventory: The skill uses the bash tool to execute training scripts (launch_training.sh), synthetic data generation (run_sdg.sh), and evaluation tools (run_eval.sh).\n
Sanitization: No evidence of sanitization or validation of the contents of the ingested datasets or specification files before they are used to drive command-line operations.

tao-generate-anomalies