The Agent Skills Directory

[SAFE]: The skill defines a structured workflow for local code architecture and implementation. It includes specific instructions to ensure all findings and artifacts remain local, and explicitly forbids agents from communicating with humans or external systems.
[SAFE]: The skill explicitly prevents unauthorized state changes by instructing the agent not to push or open pull requests without user authorization.
[SAFE]: No malicious patterns, such as prompt injection, hardcoded credentials, or obfuscated code, were detected. The skill uses standard configuration files and platform tools to achieve its goals.
[INDIRECT_PROMPT_INJECTION]: The skill has a functional surface for indirect prompt injection as it ingests untrusted data from the local repository (instructions, comments, and documentation) during the grounding phase.
Ingestion points: Repository instructions, git history, comments, and local documentation (referenced in SKILL.md).
Boundary markers: None explicitly defined to separate untrusted content from the system prompt.
Capability inventory: Local file writing (implementation phase) and execution of repository checks.
Sanitization: Not specified, though the skill instructs the agent to separate sourced intent from inference.

architect