The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to perform an extensive analysis of the current codebase to extract architectural patterns, build commands, and coding standards. This process consumes untrusted data that could contain malicious instructions designed to subvert the agent's behavior or inject unauthorized rules into the generated project configuration. * Ingestion points: Found in refs/initialization-workflow.md, where the agent is instructed to explore directory structures, read existing documentation, and map dependencies. * Boundary markers: The skill does not provide clear delimiters or 'ignore' instructions to help the agent distinguish between code/data and potential prompt injections found within the codebase. * Capability inventory: The agent uses its workspace file-reading tools to perform discovery and its file-writing tools to generate or update the CLAUDE.md file. * Sanitization: No mechanisms for filtering, validating, or escaping extracted content are specified in the workflow.

claude-md-architect