The Agent Skills Directory

[DATA_EXPOSURE]: The file references/source-evidence.md contains a hardcoded absolute local directory path: /Users/wandl/workspaces/workspace-ddd4j/workspace-ddd4j-cloud/ddd4j-cloud. This reveals the internal directory structure and the local username of the developer, which constitutes metadata leakage from the development environment.
[EXTERNAL_DOWNLOADS]: The skill's documentation in SKILL.md recommends the installation of related skills using the npx command (e.g., npx skills add full-stack-skills/ddd4j-skills). These resources trace back to the same vendor (full-stack-skills) and are documented as standard extension mechanisms for the platform.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external project data such as Maven POM files, Nacos/Sentinel configuration files, and source code to perform its observability audits.
Ingestion points: references/source-evidence.md defines the root directory for analysis, and the workflow in SKILL.md describes reading OpenSpec, POMs, and test code.
Capability inventory: The skill instructs the agent to configure Nacos registration, Sentinel rules, and probes, and to verify trace propagation across services.
Sanitization: The instructions include a specific security rule to redact sensitive Nacos/Redis credentials before any output, providing a manual safeguard against credential exposure.
Boundary markers: The instructions do not explicitly require the use of delimiters or 'ignore instructions' markers when reading external configuration files, relying on the agent's internal logic to separate data from instructions.

ddd4j-cloud-observability