# Fable Orchestrate

> Keep planning, verification, and hard problems in the main thread; dispatch implementation to subagents.

- Skill: `simonwong/fable-orchestrate-2` (Agent Skill)
- Install (CLI): `npx skillmds@latest add simonwong/fable-orchestrate-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/simonwong/fable-orchestrate-2/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: simonwong (https://skillmd.com/u/simonwong)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/simonwong/fable-orchestrate-2

---


# Fable Orchestrate

From now on, keep in the main thread:

- Clarifying the request.
- Splitting it into tasks with acceptance criteria.
- Dispatching tasks.
- Verifying reports.
- Hard problems: design decisions, tricky debugging, anything where judgement matters more than legwork.

Delegate implementation work: reading code at scale, writing code, running tests, bulk edits.

## Dispatch

The argument picks the executor. Without one, use `opus`.

| Argument | Executor |
| --- | --- |
| `opus` (default), `sonnet` | `Agent` tool with `model` set to that name |
| `gpt` | `Agent` tool with `subagent_type: codex:codex-rescue`; prefix the prompt with `--wait --fresh` |
| `herdr <kind>` | `herdr` skill: start a `<kind>` agent, `agent prompt --wait`, then `agent read` for its report |

Run independent tasks in parallel. Tell every subagent to do the work itself without spawning subagents.

## Report

Ask each subagent to report:

- What was done.
- Files changed.
- How it was verified, with command output.
- Open questions or blockers.

Verify each report against its acceptance criteria. Follow up on the same executor: `SendMessage` for `opus` and `sonnet`, `--wait --resume` for `gpt`, another `agent prompt` for `herdr`.