The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill assembles and executes Python scripts on remote servers at runtime using the ssh <server> "python3 -c ..." pattern. This is primarily used in Step 3.5 to interface with the Weights & Biases API.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external, potentially untrusted sources including remote screen logs and experiment result files.
Ingestion points: Remote screen hardcopy files and JSON result files retrieved via SSH in Step 2 and Step 3.
Boundary markers: The instructions lack delimiters or explicit boundary markers to prevent the agent from following instructions embedded in experiment logs or results.
Capability inventory: The skill is granted Bash(ssh *) and Write capabilities, which could be leveraged if the agent attempts to execute commands found in the ingested data.
Sanitization: There is no evidence of sanitization, filtering, or validation performed on the remote output before it is summarized or interpreted.
[COMMAND_EXECUTION]: The skill uses SSH to execute various commands on remote infrastructure, including screen management, file system operations (ls, cat, tail), and Python script execution.

monitor-experiment