What problem does it solve?
This Skill helps you diagnose repeated workflow failures in Claude Code and convert them into the smallest lasting improvement to your setup. It is designed for moments when a task breaks down because the current instructions, skills, hooks, agents, or integrations are missing, overloaded, or shaped incorrectly.
Core Features & Use Cases
- Root Cause Classification: Reviews conversation signals such as user corrections, rejected approaches, repeated explanations, and context overload, then maps each issue to the best architectural fix.
- Targeted Change Design: Recommends the narrowest durable change, such as updating CLAUDE.md, patching an existing skill, creating a new skill, splitting or merging skills, adding a subagent, introducing a hook, connecting MCP, or scaffolding a plugin.
- Approval-Driven Application: Structures findings with confidence, exact file targets, and rationale, then asks whether to apply changes, scaffold drafts, review one by one, or skip.
- Use Case: After a session repeatedly fails because a workflow needs deterministic execution or a specialized worker, use this Skill to identify whether the right fix is a hook, a subagent, or a refined skill instead of repeating the same correction again.
Quick Start
Ask Claude to use the reflect-architecture skill to analyze the current conversation and propose the smallest durable change for the repeated issue you just encountered.
