workflow-builder
Workflow Builder
Create and execute custom, repeatable multi-step AI workflows. Each step can use a different model, and independent steps run in parallel waves.
Capabilities
This skill has two modes:
| Mode | When to use | Reference |
|---|---|---|
| Plan | User wants to create or design a new workflow | references/plan.md |
| Run | User wants to execute, resume, or re-run a workflow | references/run.md |
Quick Start
- Create: "Create a workflow called dependency-cleanup"
- Run: "Run the dependency-cleanup workflow"
- Resume: "Resume the dependency-cleanup workflow"
Routing
Determine the user's intent from their message or $ARGUMENTS:
-
If the user says "create", "plan", "design", or "new workflow" → Read and follow
references/plan.md -
If the user says "run", "execute", "start", or "resume" → Read and follow
references/run.md -
If
$ARGUMENTSstarts withplan→ Read and followreferences/plan.mdwith the remaining argument as workflow name -
If
$ARGUMENTSstarts withrun→ Read and followreferences/run.mdwith the remaining argument as workflow name -
If unclear, ask the user: "Would you like to create a new workflow or run an existing one?"
Workflow Structure
A workflow is a directory at .somnio/workflows/<name>/ containing:
.somnio/workflows/<name>/
├── context.md — Manifest with step list and metadata
├── 01-<step-name>.md — Step 1 prompt
├── 02-<step-name>.md — Step 2 prompt
├── ...
├── config.claudecode.json — Model assignments (optional)
├── progress.json — Execution state (auto-generated)
└── outputs/ — Step output files (auto-generated)
Key Concepts
- Steps run in fresh AI contexts (separate subagents)
- Tags (
research,planning,execution) map to model assignments - Dependencies (
needs) determine parallel wave grouping - Placeholders (
{output_path},{step_N_output}) enable data flow between steps - Independent steps with no
needsrun concurrently in the same wave