The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements a lifecycle hook system (before and after generation) that supports a shell type for executing arbitrary commands.
Evidence in SKILL.md: The spec.hooks configuration allows defining commands such as git add . using type: shell.
Evidence in references/scaffold-yaml-schema.md: Confirms that kind: step hooks support the shell type.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect injection because it is designed to fetch and process external, potentially untrusted templates that can define their own shell hooks and file-writing logic.
Ingestion points: The skill fetches templates from remote sources including Git, HTTPS, S3, and OCI registries (e.g., SKILL.md mentions oci://ghcr.io/org/template:v1).
Boundary markers: The documentation provides the --skip-hooks CLI flag and ATMOS_SCAFFOLD_SKIP_HOOKS environment variable to allow users to bypass hook execution for untrusted runs.
Capability inventory: The skill has the capability to write files to the local file system and execute shell commands through hooks defined in the scaffold.yaml file (referenced in SKILL.md and references/scaffold-yaml-schema.md).
Sanitization: There is no explicit documentation regarding the sanitization or validation of template variables (answers) before they are interpolated into shell command strings in hooks.
[EXTERNAL_DOWNLOADS]: The skill downloads templates and configuration from external sources, including well-known services.
Fetches OCI-based templates from GitHub Container Registry (ghcr.io).
Supports fetching templates from Git repositories, S3 buckets, and HTTPS URLs.
[DYNAMIC_EXECUTION]: The skill utilizes Go templates (Gomplate/Sprig) and Common Expression Language (CEL) for dynamic logic and content rendering.
Uses CEL for conditional logic in when: fields for both questionnaire prompts and file generation.
Uses Go templates for rendering file content, dynamic matrix generation paths, and computing field options.

atmos-scaffold