normalize-agent-for-claude

To reduce prompt bloat in an AGENT.md file, audit and normalize it against the Claude Code runtime to separate role identity from procedural content. This process moves oversized workflow instructions into appropriate skills or references while preserving the agent's core expertise.

Migrating a custom agent to the Claude Code runtime involves auditing your local AGENT.md file against runtime behaviors like prompt loading and tool controls. You generate synchronized Markdown and JSON DAG plans for controlled rewrites without introducing unsupported configurations.

DAG rewrite planning for Claude agent normalization produces synchronized Markdown and JSON artifacts covering findings, dependencies, guardrails, and file changes. It maps out validated modifications before performing a controlled rewrite of the AGENT.md file.

Yes, you can use plan mode to audit your AGENT.md file before rewriting it. The skill evaluates the agent against Claude Code runtime behavior, including frontmatter and skill preloading, and produces a validated rewrite plan without blindly modifying role-specific instructions.

Duplicated policy and oversized workflow instructions in a Claude agent occur when procedural content is not properly separated from role identity. Auditing the AGENT.md file against the Claude Code runtime identifies these redundancies and moves procedures into appropriate skills.

You should avoid automated rewrites when you need to strictly preserve an agent's role identity and expertise without introducing unsupported configurations. Controlled normalization requires reading the Claude agent runtime reference and generating DAG plans before performing any requested file changes.