# Revisit RP contextbuilder + chatsend for reviews

> PR #73 migrated from RP's --response-type review back to manual prompt building because the two-step approach was too brittle.

- Skill: `tools-only/revisit-rp-contextbuilder-plus-chatsend-for-reviews` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add tools-only/revisit-rp-contextbuilder-plus-chatsend-for-reviews`
- Raw SKILL.md: https://api.skillmd.com/api/skills/tools-only/revisit-rp-contextbuilder-plus-chatsend-for-reviews/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: tools-only (https://skillmd.com/u/tools-only)
- Updated: 2026-09-29
- Page: https://skillmd.com/skills/tools-only/revisit-rp-contextbuilder-plus-chatsend-for-reviews

---

# Revisit RP context_builder + chat_send for reviews

## Background

PR #73 migrated from RP's `--response-type review` back to manual prompt building because the two-step approach was too brittle.

## Current approach (manual prompt building)
1. `setup-review` → context building
2. `prompt-get` → get builder's handoff
3. Build custom prompt with explicit verdict tag requirement
4. `chat-send` → single message with our prompt

## Alternative tested (2026-01-26)

Tested `context_builder` (no response_type) → `chat_send` with verdict in message:
- Discovery worked: builder selected relevant files
- Verdict came back when included in chat message
- BUT: Previously tried and found brittle

## Why it was brittle (from PR #73)
- RP's `--response-type review` returns its own verdict format (request-changes, approve)
- Required follow-up message to get `<verdict>...</verdict>` tags
- Timing/reliability issues

## Eric's suggestion (pvncher)
- "You can edit the review prompt" in RP settings
- "Prompt your way there" - add XML tags in instructions
- Problem: Can't require all users to change RP settings

## Tasks when revisited
1. Test `context_builder` (no response_type) → `chat_send --mode review` with verdict in message
2. Verify reliability over multiple runs
3. Compare token usage vs manual approach
4. Check if RP has programmatic way to set review prompt defaults