The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill processes untrusted data from external Weights & Biases runs which could contain malicious instructions meant to influence the LLM's interpretation of the status reports.
Ingestion points: Data enters the context via wandb.Api().runs() in scripts/check_runs.py and scripts/watch_runs.py, and wandb.Api().run() in scripts/run_details.py.
Boundary markers: Absent. External data (run names, tags, config keys) is interpolated directly into the printed output without delimiters.
Capability inventory: The skill's scripts are restricted to read-only API calls and printing to standard output. No file-system writes or dangerous subprocess executions are triggered by the fetched data.
Sanitization: None. External strings are printed as-is.
Data Exposure (SAFE): The skill requires a WANDB_API_KEY, but correctly instructs the user to set this via environment variables or the official wandb login CLI, avoiding hardcoded secrets.
External Downloads (SAFE): The skill depends on the wandb Python library, which is a standard, trusted package for machine learning experiment tracking.

wandb