The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides instructions for triaging and debugging incidents using external data sources such as Cloud Manager logs, Splunk query results, and Jira tickets. This creates a surface for indirect prompt injection if malicious instructions are embedded in the logs or tickets processed by the agent.
Ingestion points: workflow-triaging/SKILL.md describes gathering data from environment logs, Splunk queries, and ticket descriptions to classify symptoms.
Boundary markers: There are no explicit instructions or delimiters defined to separate user-provided log data from agent instructions.
Capability inventory: The skill allows for the generation of Java code for OSGi services, Maven configurations for deployment, and provides curl command templates for repository interaction (workflow-development/SKILL.md, workflow-triggering/SKILL.md).
Sanitization: The skill does not provide guidance on sanitizing or escaping log content before processing.
[COMMAND_EXECUTION]: The documentation includes multiple curl command templates for interacting with the AEM repository and services. While these examples target localhost:4502 and are intended for development and verification, they represent a capability for the agent to execute shell commands to interact with the environment (workflow-triggering/SKILL.md, workflow-debugging/reference.md).
[CREDENTIALS_UNSAFE]: The skill includes hardcoded default credentials (admin:admin) within several curl command examples targeting localhost:4502. While these are the standard factory defaults for local AEM development environments and serve as placeholders, they are technically hardcoded credentials (workflow-triggering/SKILL.md, workflow-launchers/SKILL.md).

aem-workflow