The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various shell and Python scripts to verify the integrity of the scaffolded repository and configure hosting settings. This includes running scripts like check-identity.sh, check-reply.py, and check-host.py. The latter uses a user-provided admin token to configure branch protection and default settings, which is performed only after explicit user confirmation.
[DYNAMIC_EXECUTION]: During the initialization process, the skill dynamically generates and updates the CLAUDE.md repository configuration file. It uses a shell script snippet to concatenate a standard pointer block with any existing content, ensuring the repository adheres to the required standards from its creation.
[INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface for indirect prompt injection by reading configuration assets and reference documentation from the local environment (~/.claude/skills/atelier/) and incorporating them into the new project.
Ingestion points: Reads project templates, gate scripts, and architectural references from the local file system and target directory.
Boundary markers: Interaction is governed by strict discipline rules (e.g., Rules 24-25), though specific data delimiters are not explicitly defined in the scaffolded content.
Capability inventory: The skill has the ability to execute bash and Python scripts, perform git operations, manage package dependencies via bun and mvnw, and write files.
Sanitization: Implements an IDENTITY_DENYLIST mechanism to scrub sensitive personal or corporate identifiers from the generated codebase before the initial commit.

atelier-greenfield