The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves and analyzes untrusted external data from pod logs and system events during job diagnostics. This data is produced in external training environments and could potentially contain malicious instructions intended to manipulate the agent's reasoning.\n
Ingestion points: The get-pod-logs and get-job-events commands in SKILL.md fetch logs and events from the distributed training environment.\n
Boundary markers: The skill instructions do not specify any delimiters or safety prompts to isolate untrusted log data from agent instructions.\n
Capability inventory: The skill leverages the Aliyun CLI to perform various read operations and environment configurations.\n
Sanitization: There is no evidence of filtering or sanitization of the retrieved logs before the agent processes them.\n- [EXTERNAL_DOWNLOADS]: The skill provides instructions for downloading the Aliyun CLI binary and associated plugins from official vendor sources.\n
CLI binaries are downloaded from the official vendor CDN at aliyuncli.alicdn.com (referenced in references/cli-installation-guide.md).\n
Official plugins such as aliyun-cli-paistudio are installed via the Aliyun CLI plugin manager.\n
The skill suggests the use of npx to add official diagnostic tools from the vendor's skill repository.

alibabacloud-pai-dlc-job-diagnostics