Prompt templates

A prompt template is a named, reusable prompt body. If you find yourself pasting the same instructions into a dozen workflows, lift them into a template and reference the template by name. The template is the single source of truth; every workflow that uses it stays in sync when you edit one file.

Think of templates as the spice rack next to the recipe book. The "summarize decisions from a transcript" instruction set isn't a full recipe on its own, but every meeting workflow uses it. Move it into a template once and every workflow gets the better wording the next time you improve it.

Templates vs. workflows

The simple test:

• If the thing is a full sequence with multiple steps, dependencies, and an output that lands somewhere, it's a workflow.
• If the thing is a single prompt body you keep using inside workflows, it's a template.

Templates are workflow primitives, not standalone executables. You don't zen workflow run a template; you reference one from inside a workflow node.

Where to start

Pull any prompt you've pasted into a workflow more than twice. Move it into a template file. Reference the template from the workflows that need it. Edit the template the next time the instructions need a tweak; every workflow picks up the change at next fire.

Advanced: the file format and call syntax

What a template looks like

Templates live in .zen/commands/ (or ~/.zen/commands/ for templates that should apply to every project on your machine). Each template is a .md file with a frontmatter block:

---
description: Extract decisions from a meeting transcript.
argument-hint: <transcript file path or content>
---

You're reading a meeting transcript. Pull out only the decisions that were
made, in the form "We decided to X because Y." Skip discussion that didn't
end in a decision. Group decisions by topic if there are more than five.

Transcript:
$ARGUMENTS

The filename (without extension) is the template name. summarize-decisions.md registers as summarize-decisions.

Calling a template from a workflow node

nodes:
  - id: pull-transcript
    type: bash
    command: cat $TRANSCRIPT_FILE

  - id: extract-decisions
    type: prompt
    template: summarize-decisions
    depends_on: [pull-transcript]
    vars:
      TRANSCRIPT: $pull-transcript.output

The node's prompt becomes the template body with variables filled in. You can still set per-node settings (provider, timeout, retries); they apply on top of the template.

Overriding parts of a template

You can pass extra instructions per-call without forking the template:

  - id: extract-decisions
    type: prompt
    template: summarize-decisions
    extra_instructions: |
      This was a board meeting; focus on capital and strategy decisions only.
    vars:
      TRANSCRIPT: $pull-transcript.output

extra_instructions: is appended to the template body before the model sees it.

Discovery

zen command list prints the templates Z.E.N. knows about, from both .zen/commands/ and ~/.zen/commands/. Project templates win over global ones with the same name.