The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses standard shell commands like ls, cat, diff, find, and rm to verify project files and manage training artifacts. These commands are part of the intended diagnostic and cleanup workflow.
[CREDENTIALS_UNSAFE]: The skill references environment variables used by experiment trackers (such as WANDB_API_KEY) to confirm their presence. It does not hardcode, store, or exfiltrate any secret values.
[DATA_EXFILTRATION]: The skill accesses local configuration and dataset paths for verification purposes. There are no patterns indicating that sensitive data is being transmitted to external locations.

launch