The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to execute ssh commands. It dynamically constructs shell commands using user-provided arguments (e.g., <server>, <name>) and internal placeholders, which presents a surface for command injection if the underlying tool or agent does not properly sanitize the input.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and interprets untrusted data from remote sources, including experiment logs, screen session hardcopies, and JSON result files.
Ingestion points: ssh outputs from tail, cat, and screen, as well as data retrieved via the Weights & Biases API.
Boundary markers: Absent; there are no instructions to the agent to ignore or delimit embedded instructions in the logs.
Capability inventory: The agent has access to Bash (with ssh and echo), Read, Write, and Edit tools.
Sanitization: Absent; the skill does not specify any filtering or validation of the ingested log content before the 'Interpret' or 'Summarize' steps.
[DATA_EXFILTRATION]: The skill accesses local configuration files including vast-instances.json and ~/.claude/feishu.json. While these are used for functional requirements like cost reporting and sending notifications, these files may contain sensitive metadata or authentication tokens.

monitor-experiment