harness-prompt

To keep system prompts in sync across Claude, Codex, and Copilot, maintain a single canonical source file and render it into provider-native instruction files. This prevents configuration drift by generating CLAUDE.md, AGENTS.md, and copilot-instructions.md from one centralized prompt.

Managing a canonical system prompt for multiple AI coding agents involves centralizing the source text and applying it to generate provider-specific configurations. You can scaffold the prompt once and propagate it to generate the respective CLAUDE.md and AGENTS.md files automatically.

Yes, you can apply per-provider overrides to Copilot and Codex instruction files using sidecar YAML configurations. This allows you to maintain the canonical system prompt while injecting specific instructions for individual AI providers during the rendering process.

To scaffold and generate CLAUDE.md and copilot-instructions.md, run a prompt scaffolding command to create the canonical source, edit the markdown file, and execute an apply command to render the provider-native instruction files into your agent-harness workspace.

System prompt drift occurs when switching between different AI coding providers because instruction files are maintained independently. Centralizing the source prompt and applying it to generate CLAUDE.md, AGENTS.md, and copilot-instructions.md prevents this drift.