respond

The respond skill serves as the manual interface bridge for the Dual Agent development environment, allowing for the reliable delivery of findings from secondary agents (like Codex or ChatGPT) back to Claude. It is primarily used as a fallback mechanism when the automated read-task flow fails to capture output or when a user needs to curate, modify, or finalize a response before it is retrieved by the primary agent.

• Facilitates manual transmission of analysis, code, or project status updates between agent environments.
• Provides a structured format for responses, including Task Type, timestamp, summary, and detailed output, ensuring compatibility with the Codex CLI and Claude Code workflows.
• Resolves complex directory structures, automatically locating the .agent-collab path even when commands are executed from deep within nested subdirectories.
• Updates the project status file to signal that new information is available for retrieval.

Designed for developers working in split-pane tmux sessions or those managing asynchronous AI tasks, this skill ensures that context remains consistent across multiple AI models. Use it when you need to bridge the gap between agents, specifically if the read-task process does not auto-complete as expected. It ensures the environment state is correctly updated, allowing Claude to resume operations using the /codex-read command. Practical inputs include your finalized code blocks, security findings, or architectural summaries, which are written into a standardized markdown response file.