Interactive 8-node DAG workflow turning a raw idea into BKM-format Work Orders through four AI phases with human approval gates between each: understand-idea -> foundation-gate scope-and-approach -> approach-gate risk-and-decomposition -> decomposition-gate generate-wos -> review-wos Originally authored by Robert (Bob) L O'Donnell (@lamachine) in Archon PR #1647 (https://github.com/coleam00/Archon/pull/1647) as a bundled default workflow. Repackaged here as a standalone external repo so it can be SHA-pinned and published through the Archon Workflow Marketplace. |
||
|---|---|---|
| .archon | ||
| LICENSE | ||
| README.md | ||
archon-idea-to-wo
An interactive Archon workflow that turns a raw idea into a set of self-contained, implementation-ready Work Orders (WOs) — through four AI phases with human approval gates between each.
Published in the Archon Workflow Marketplace.
Credit
Originally authored by Robert (Bob) L O'Donnell in Archon PR #1647, where it was proposed as a bundled default workflow. With the launch of the community marketplace, the workflow was reshaped into a standalone, SHA-pinned external repo so other Archon users can install it with one command instead of waiting for a release. The original design, prompts, and Work Order template are Bob's; this repo is a packaging adaptation.
What It Does
Most autonomous coding loops fail because the input is under-specified. This workflow forces the missing thinking up front: foundation questions → scope research → risk-aware decomposition → BKM-format Work Order files. The output is a directory of WO-NNN-*.wo.md files that the user can hand straight to archon-piv-loop (or any other implementer).
Flow
archon workflow run archon-idea-to-wo "add a dark mode toggle"
│
▼
[understand-idea] AI reads the codebase, asks 5 foundation questions
│
▼ ━━━ approval gate ━━━ User answers
│
[scope-and-approach] AI researches deeply, proposes scope + alternatives
│
▼ ━━━ approval gate ━━━ User approves or corrects
│
[risk-and-decomposition] AI orders the work, sizes WOs, calls out risks
│
▼ ━━━ approval gate ━━━ User approves the breakdown
│
[generate-wos] AI writes BKM Work Orders to $ARTIFACTS_DIR/work-orders/
│
▼ ━━━ approval gate ━━━ User signs off
Each phase reads the prior gate's captured user response, so corrections actually propagate.
Install
From inside any git repo where you've installed Archon:
archon workflow install archon-idea-to-wo
This pulls the workflow YAML and the four command files at the pinned SHA from the marketplace entry into your local .archon/.
Run
archon workflow run archon-idea-to-wo "the feature, fix, or capability you want to plan"
The workflow is interactive — approval gates pause for input in the Archon Web UI (or CLI prompt). Final Work Orders land in $ARTIFACTS_DIR/work-orders/ along with a README.md index.
Output Format — BKM Work Orders
Each generated WO-NNN-*.wo.md includes:
- Priority + effort + domain + Conductor track
- Context (why this WO exists, how it fits the plan)
- "Files to Read First" table (verified, with line ranges)
- Numbered task steps, with explicit decision points
- Anti-patterns to avoid
- Test cases to write
- Definition of Done with validation commands
The template is designed so each WO is autonomously implementable — no additional context required.
When to Use This Workflow
Good fit:
- New feature where you have an idea but not a spec
- Multi-step refactor that needs ordering
- Backlog item that needs to be broken into reviewable pieces before implementation
Not a fit:
- PRD creation — use
archon-interactive-prd - Direct implementation when scope is already clear — use
archon-piv-loop - Single-file bug fixes — overkill
Repository Layout
.archon/
├── archon-idea-to-wo.yaml # main workflow (8 nodes)
└── commands/
├── archon-wo-understand-idea.md
├── archon-wo-scope-and-approach.md
├── archon-wo-risk-and-decomposition.md
└── archon-wo-generate-wos.md
This is the canonical layout for marketplace directory-format submissions. The Archon installer expects <slug>.yaml at .archon/ root and walks commands/, scripts/, and skills/ one level deep.
License
MIT — see LICENSE.