The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill uses joblib.load in the active_learning_pipeline.py script to load a pre-trained model.\n
Evidence: model = joblib.load("models/current_model.pkl") in references/EXAMPLES.md.\n
Description: Loading models with joblib or pickle can lead to arbitrary code execution if the source file is untrusted or malicious.\n\n- [EXTERNAL_DOWNLOADS]: The skill downloads the Label Studio software and associated Docker images.\n
Evidence: pip install label-studio and docker pull heartexlabs/label-studio:latest in SKILL.md.\n
Description: These downloads originate from the official repositories of the Label Studio project.\n\n- [PROMPT_INJECTION]: The skill ingests external datasets for the purpose of labeling, which creates an attack surface for indirect prompt injection.\n
Ingestion points: Loading data from data/unlabeled_texts.csv and exports/annotations.json in references/EXAMPLES.md.\n
Boundary markers: No specific delimiters or safety instructions are used to separate data from system prompts.\n
Capability inventory: The skill is granted Bash and file management tools (Read, Write, Edit) in SKILL.md.\n
Sanitization: The provided code does not perform sanitization or validation of the input text content before it is processed or displayed in the labeling interface.

label-training-data