gmickel

flow-next-impl-review

@gmickel/flow-next-impl-review
gmickel
390
21 forks
Updated 1/18/2026
View on GitHub

John Carmack-level implementation review via RepoPrompt or Codex. Use when reviewing code changes, PRs, or implementations. Triggers on /flow-next:impl-review.

Installation

$skills install @gmickel/flow-next-impl-review
Claude Code
Cursor
Copilot
Codex
Antigravity

Details

Pathplugins/flow-next/skills/flow-next-impl-review/SKILL.md
Branchmain
Scoped Name@gmickel/flow-next-impl-review

Usage

After installing, this skill will be available to your AI coding assistant.

Verify installation:

skills list

Skill Instructions


name: flow-next-impl-review description: John Carmack-level implementation review via rp-cli for current branch changes. Use when reviewing code changes, PRs, or implementations. Triggers on /flow-next:impl-review.

Implementation Review Mode (CLI)

Conduct a John Carmack-level review of implementation changes on the current branch using RepoPrompt's context builder and chat.

Role: Code Review Coordinator (NOT the reviewer) Tool: rp-cli for context building and chat delegation

Input

Arguments: #$ARGUMENTS Format: [additional context, focus areas, or special instructions]

Example: /flow-next:impl-review focus on the auth changes, ignore styling

Reviews all changes on the current branch vs main/master.

Setup

FLOWCTL="${CLAUDE_PLUGIN_ROOT}/scripts/flowctl"

FIRST: Determine Review Mode

Check: which rp-cli >/dev/null 2>&1 If NOT available: inform user rp-cli is required for this skill.

Option Parsing (skip questions if found in arguments)

Parse the arguments for these patterns. If found, use them and skip the question:

Review mode:

  • --mode=rp or --rp or "rp chat" or "repoprompt" → RepoPrompt chat
  • --mode=export or --export or "export" or "external llm" → export for external LLM

If options NOT found

If review mode was already chosen earlier in this conversation (e.g., user answered "2a" or "2b" during /flow-next:plan or /flow-next:work setup): → Use that mode, don't ask again.

If invoked directly without prior context, output this text (do NOT use AskUserQuestion tool):

Both modes use RepoPrompt for context building (builder, file selection, codemaps).
The difference is where the review happens:

Review mode:
a) RepoPrompt chat (default) — review via rp-cli chat
b) Export for external LLM — export context file for ChatGPT, Claude web, etc.

(Reply: "a", "b", "export", or just tell me)

Wait for user response. Parse naturally.

Context Sources

The workflow gathers context from:

  • Git diff and commit messages
  • Epic/task specs if a Flow ID is known
  • PRD/architecture docs

If you know which Flow epic/task this work relates to, include context:

$FLOWCTL show <id> --json
$FLOWCTL cat <id>

Critical Requirement

DO NOT REVIEW CODE YOURSELF – you are a coordinator, not the reviewer.

Your job is to:

  1. Use rp-cli -e 'windows' to find the RepoPrompt window
  2. Use rp-cli -w <id> -e 'builder ...' to build context
  3. Use rp-cli -w <id> -e 'chat ...' to execute the review

The RepoPrompt chat conducts the actual review with full file context.

Workflow

Read workflow.md and follow each phase in order. Phases include change identification, context building, and review execution.

rp-cli Reference

Read rp-cli-reference.md for command syntax and examples.