RepoPrompt Subagent

Overview

Delegate tasks to a RepoPrompt chat session with curated file context. Use this skill to prepare context, send a clear task prompt, and iterate until the delegated task is complete.

Trigger examples

• "Spawn a RepoPrompt sub-agent to find where X is defined."
• "Delegate this code search/plan/review to a sub-agent."
• "Have RepoPrompt explore the repo and report back."
• "Ask a sub-agent to draft the changes."

Workflow

1. Clarify the delegated task and expected output.
2. Build context with RepoPrompt tools.
   • Use get_file_tree to orient yourself.
   • Use file_search to find entry points.
   • Use manage_selection with mode="full" for key files and mode="codemap_only" for broad scans.
   • Use read_file for targeted sections.
3. Start the sub-agent with chat_send (sometimes referred to as send_chat).
   • Set new_chat: true.
   • Set chat_name and mode (plan, chat, edit, review).
   • Provide selected_paths when you want to override current selection.
4. Iterate on the same sub-agent session.
   • Capture chat_id and reuse it with chat_send for follow-ups.
   • Use review mode when you need diff-focused feedback.

Sub-agent directives to include in the prompt

• Follow AGENTS.md and repo conventions as canonical.
• Keep scope tight; avoid unrelated refactors.
• Use RepoPrompt tools for discovery; do not assume file contents.
• Favor correctness and recoverability over cleverness.
• Call out conflicting instructions explicitly.
• Prefer more context over too little; avoid huge files unless needed.
• Ask 1-3 clarifying questions if ambiguous.
• Start with the direct answer or plan.
• Provide file path references using inline code (e.g., lib/kimojo/foo.ex).
• Call out risks, missing tests, or open questions.
• Remember limitations: no shell commands/tests; only selected files; git diffs only in review mode.

Prompt template

<task>
Describe the delegated goal in 1-3 sentences.
</task>
<context>
List constraints, important files, and expected output format.
</context>
<constraints>
- Follow AGENTS.md and repo conventions.
- Keep scope tight; avoid unrelated refactors.
- Ask clarifying questions if needed.
</constraints>

Example

<task>
Find where supported agents are defined and how the UI should mirror them.
</task>
<context>
Use RepoPrompt tools to locate the backend source of truth and relevant UI files.
</context>