The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to download and install official IBM Cloud CLI tools and third-party monitoring agents. These resources are fetched from well-known domains including clis.cloud.ibm.com, ibm.biz, assets.us-south.logging.cloud.ibm.com, and the official Sysdig (Draios) GitHub repository. These are recognized sources within the IBM Cloud ecosystem.
[COMMAND_EXECUTION]: Includes a large volume of shell commands for infrastructure management using ibmcloud, kubectl, oc, and terraform. These commands are standard for cloud operations and administrative tasks.
[INDIRECT_PROMPT_INJECTION]: The skill describes procedures for exporting and analyzing cloud activity logs and application logs (e.g., ibmcloud at export). These logs constitute an external, untrusted data source that could theoretically be used for indirect prompt injection if they contain malicious instructions designed to influence the agent's behavior during log analysis.
Ingestion points: Audit logs and application logs accessed via IBM Cloud Activity Tracker and Log Analysis tools.
Boundary markers: The provided command templates do not include explicit delimiters or warnings to ignore instructions embedded in the logs.
Capability inventory: The agent is granted capabilities to execute cloud management commands and interact with infrastructure through various CLI tools.
Sanitization: The skill provides templates for standard operations but does not include specific sanitization or filtering logic for the log data being processed.

ibm-cloud