rp-build

rp-build

Use this when the user wants code changes and the task benefits from curated context.

Check workspace binding first. On mismatch, pass root=<absolute path> on every rp-mini tool call. If a shell is available and no MCP client is loaded, use node packages/server/dist/cli.js tool <workspace> <tool> --json-args '...' or its wrappers.

1. Verify the workspace is the intended repo and make a quick scan with get_file_tree, file_search, or get_code_structure.
2. Clarify first: if task scope, target behavior, or acceptance criteria are ambiguous, ask the user with AskUserQuestion before spawning the builder.
3. Spawn the context-builder subagent with:
   • task: reformulated user request with codebase terms from the quick scan
   • budget: default 120k for plan work unless caller supplied a hard budget
   • response_type: plan
   • enhancement mode: rewrite unless the user requested augment or preserve
4. Question loop: this is the rp-build/rp-investigate question loop. If the builder returns <questions> plus a saved profile name, read the cited files, check git history or searches, and answer only conclusive questions with <answer key="..." source="orchestrator" >...</answer>. Escalate with AskUserQuestion only for blocking AND high-stakes questions: irreversible/destructive actions, product policy, money/auth/data-loss. Append <answers> with prompt op=append, resume with manage_selection op=load_profile, and remember advisory questions never interrupt.
5. Act on the handoff: implement directly against the curated selection and plan. Use existing repo patterns and avoid unrelated refactors.
6. Validate with relevant tests, lint/format, or build commands for the touched area.
7. Report changes, validation results, assumptions, and remaining risks.

Do not skip context-builder for non-trivial code work; the quick scan is orientation, not the deep exploration.