The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a scaffolding engine that interpolates project metadata and context variables into file templates, creating a surface for indirect prompt injection.
Ingestion points: Input variables such as directory names (targetDir) and other fields in the CommandContext are interpolated into files like package.json and pyproject.toml as shown in references/command-interface.md.
Capability inventory: The writeFile and createDirectory methods in the Command base class provide the ability to write arbitrary content and structures to the local file system.
Boundary markers: No specific delimiters or safety instructions are provided in the template string examples to isolate interpolated content from the surrounding file structure.
Sanitization: The reference examples do not demonstrate explicit sanitization or validation of context variables before they are written to the disk.
[EXTERNAL_DOWNLOADS]: The skill references external templates for project scaffolding.
The RunCopierTemplate logic resolves and fetches project templates from the author's GitHub repository at https://github.com/delorenj/hermes-agent-template.
[COMMAND_EXECUTION]: The skill documentation includes instructions for running command-line tools to test and initialize subsystems.
Development and testing involve executing the local CLI source using the bun runtime.
Recipe definitions suggest that users run initialization commands such as bun install, mise install, or docker-compose after the scaffolding process is complete.

pjangler