The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources including project POM files and toolchain command outputs.
Ingestion points: Workflow steps in SKILL.md and references/workflow.md instruct the agent to read the root POM model and capture output from git branch, java -version, and ./mvnw -version.
Boundary markers: The instructions lack specific delimiters or instructions to ignore potential malicious prompts embedded within these external files or command outputs.
Capability inventory: The skill utilizes command execution capabilities via git, java, and mvnw subprocesses.
Sanitization: There is no evidence of input validation, escaping, or filtering for data ingested from the project workspace.
[CREDENTIALS_UNSAFE]: The file references/source-evidence.md contains a hardcoded absolute file path (/Users/wandl/workspaces/workspace-ddd4j/...) which exposes a local username and directory structure, potentially leaking environment details.

ddd4j-quarkus-version-selection