# Recursive Implement

> Build a project using recursive multi-agent development with TDD

- Skill: `jartan-llc/recursive-implement` (Agent Skill)
- Install (CLI): `npx skillmds@latest add jartan-llc/recursive-implement`
- Raw SKILL.md: https://api.skillmd.com/api/skills/jartan-llc/recursive-implement/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: Jartan-LLC (https://skillmd.com/u/jartan-llc)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/jartan-llc/recursive-implement

---


# Recursive Implement

Build a project (or feature) using waves of parallel nested subagents with TDD at every level.

## Process

### 1. Gather Context

Collect everything the orchestrator needs into a single prompt:

- **From the user's request:** the project description, issue number, or plan file path
- **If issue number:** fetch the issue body with `gh issue view <number>` and include it
- **From conversation:** if the user discussed requirements, architecture decisions, constraints, or preferences in this session, summarize the relevant parts
- **From existing project:** if working in an existing codebase, note the language, test framework, key conventions, and any relevant file paths

### 2. Launch Orchestrator

Spawn a `recursive-orchestrator`. The prompt should contain:

```text
[Project description or plan file path]

[Conversation context if any -- requirements, constraints, decisions made]

[Existing project context if any -- language, conventions, key files]

Working directory: [cwd]
```

### 3. Report Results

Relay the orchestrator's final report to the user.

## Failure Recovery

If the orchestrator reports a failure, present the context to the user. On re-run:

- The wave plan persists in `.claude/workspace/` -- pass the file path to skip re-planning
- Completed source files and tests remain on disk -- the orchestrator can build on them