The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements a hook system (spec.hooks) that allows for the execution of arbitrary shell commands (type: shell) during the scaffolding process. The documentation includes examples such as git add . being executed as a post-generation step.
[EXTERNAL_DOWNLOADS]: The engine supports fetching template configurations and files from external remote sources, including Git repositories, HTTPS URLs, S3 buckets, and OCI registries. These resources are fetched at runtime based on the template selection.
[INDIRECT_PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it ingests and processes untrusted template data from remote sources which then drives file-system writes and command execution.
Ingestion points: Remote template files (scaffold.yaml) and source files fetched from Git, OCI, S3, or HTTPS sources.
Boundary markers: The system uses Go template delimiters (default {{ }}) and supports custom delimiters (e.g., [[ ]]) to isolate template logic.
Capability inventory: The skill possesses capabilities for arbitrary file writing, directory traversal via globs, and shell command execution via the lifecycle hook system.
Sanitization: Content is rendered via a template engine and logic is gated by Common Expression Language (CEL), but the hook system executes raw command strings provided in the template configuration.
[DYNAMIC_EXECUTION]: The skill utilizes Go-template expressions and Common Expression Language (CEL) to dynamically evaluate questionnaire logic (when:), file paths (target:), and matrix generation axes at runtime. Computed fields also allow for the derivation of values from other user-supplied answers using template expressions.

atmos-scaffold