---
name: task-orchestration-b534e0e5bd97
description: Decomposing work into parallel subagent tasks using native Workflow/Task tooling and the Write-Review-Fix-Confirm (WRFC) template. No background daemon, no runtime engine. Load when planning how to split work across multiple agents.
---

# task-orchestration

v1's orchestration text predates native multi-agent workflow support and was written around a
bespoke runtime engine (a background daemon coordinating agents via IPC, directives, and a
phase-machine). That daemon is cut for v2. The deep review found no standing usage that
justified keeping a background process running per session, and native Task/Workflow tooling
already covers the observed needs. This skill is the rebuild: orchestration using what the
platform gives you directly, plus the WRFC template as a pattern, not infrastructure.

## Decomposing work

Break a task into steps where each step:
- Has a single agent type responsible for it (engineer, tester, refutation-reviewer, architect).
- Has an explicit "done" condition, not a vague direction ("implement the API" is not a step;
  "implement POST /api/posts per the spec architect produced, with the validation and error
  cases it lists" is).
- States its real dependencies: which steps must finish first, and what output they hand off.

Independent steps (no shared dependency) can run as parallel Task-tool calls in the same turn.
Dependent steps run sequentially, with each agent's output feeding the next agent's task
description directly. Don't route it through a shared mutable file unless the next agent
actually needs to read that file itself.

## The WRFC pattern: Write → Review → Fix → Confirm

For any change worth reviewing before it ships:

1. **Write.** The engineer agent implements the change.
2. **Review.** The refutation-reviewer agent reviews the actual diff (see the `review-scoring`
   skill for the rubric: a defect list with severity, not a scalar score).
3. **Fix.** If the review found `CONFIRMED critical`/`high` defects, the engineer addresses
   them. Cap this at a small, explicit number of iterations (2-3). An unbounded fix loop is a
   sign the task needs re-scoping, not more iterations.
4. **Confirm.** The FIX gets reviewed too, not assumed correct. Only after a fix iteration
   passes review (or the loop is explicitly capped and escalated to the user) is the change done.

This is a workflow *pattern* you run in-band as the orchestrator: spawning the engineer, then
the reviewer, then (conditionally) another engineer pass, then a final reviewer pass, using
native Task calls. It needs no daemon, no directive-delivery IPC, and no phase-machine state
file. The orchestrating session's own turn sequence *is* the state machine.

`/goodvibes:codebase-review` is the packaged entry point for this pattern applied to a
whole diff/PR. See that command for the concrete workflow.

## When not to use the full loop

A one-line typo fix or a change with no meaningful failure surface doesn't need a dedicated
review agent. Use judgment. WRFC earns its cost on changes where "did this actually work"
is a real question, not on everything unconditionally (v1's automatic-review-on-every-agent
behavior was one of the field-identified sources of wasted review cycles on read-only work).


---
Source attribution: mgd34msu
Original project: https://github.com/mgd34msu/goodvibes-plugin/blob/d19fa3c57eda377c032d3cd3e897fe127d1e99aa/plugins/goodvibes/skills/task-orchestration/SKILL.md
License: MIT
License information: https://opensource.org/license/mit
Original upstream instructions; downloaded from Mega AI Hub.

## Upstream license notice

MIT License

Copyright (c) 2025 GoodVibes Contributors

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.



# Mega AI Hub resource metadata

Creator / publisher: mgd34msu
Original project: https://github.com/mgd34msu/goodvibes-plugin/blob/d19fa3c57eda377c032d3cd3e897fe127d1e99aa/plugins/goodvibes/skills/task-orchestration/SKILL.md
Attribution basis: Source repository record

Fields of work: Product Management, Project & Program Management, Development & Engineering
Roles: Product Manager, Project Manager, Backend Developer, Frontend Developer, DevOps / SRE
Specialties: Frontend, Design, React, Word Documents, API, Database, Testing, DevOps, Backend
Tools: Tool not specified
Country applicability: Country not specified
Classification basis: Source category, role and specialization labels
Instruction SHA-256: 78b8c78251dbdd840cea5aee6b28f925b35c70f89cc865a4f46aab067f82e36c