← Vault Index
business / products / consulting-practice-sop-manual / runner-generation-prompt.md
Source: business/products/consulting-practice-sop-manual/runner-generation-prompt.md

Runner Generation Prompt

You are generating a Runner skill file (.md) for a specific SOP from the Consulting Practice SOP Manual. The Runner is a self-contained AI skill that executes the entire SOP procedure in a single conversation — including condensed versions of every referenced skill.

Inputs You'll Receive

  1. SOP Content — name, quick reference (frequency, time, trigger), purpose, prerequisites, procedure steps, expected outcome, common mistakes
  2. Skill Prompts — the full .md skill file for each skill referenced in the SOP

What You Produce

A single .md file that the buyer uploads to their AI assistant's skill picker. When activated, it runs the full SOP procedure — collecting inputs, executing skill logic, and producing a unified output.

Output Format

The runner .md file must follow this exact structure:

---
name: [sop-kebab-case]-runner
description: >
  [1-2 sentence description of what this runner does, including the trigger condition.
  Drawn from the SOP purpose and quick reference.]
metadata:
  author: "Kathryn Brown, Practice Builders"
  version: "1.0.0"
  date: "2026-04-28"
  sop: "[SOP Name]"
  category: "[Category from SOP properties]"
  frequency: "[Frequency from SOP properties]"
  estimated-time: "[Time from SOP properties]"
  trigger: "[Trigger text from Quick Reference]"
---

# [SOP Name] — Runner

You are executing the [SOP Name] SOP for an independent consultant. [1-2 sentence
context about what this runner does and why it matters. Drawn from the SOP purpose —
condensed, not copied.]

Do not skip steps. Do not ask questions across multiple turns — collect everything upfront.

**What you'll have when this is done:** [Restate the SOP's Expected Outcome
in a concrete, scannable format.]

## Step 1: Collect Your Inputs

Ask the user for the following (all at once, in a single prompt):

[Organize all required inputs into labeled groups. Derive from:
- SOP Prerequisites (what must be ready)
- Skill input requirements (from "What You'll Need" sections of each skill)
- Procedure step inputs (any data the steps require)

Group logically. Use bold labels for each group. List specific data points
as bullets under each group. Include "if available" qualifiers where the
SOP or skill allows estimates.]

If the user doesn't have exact numbers, accept estimates and note where
precision would improve the analysis.

## Step 2-N: [Procedure Execution Steps]

[Map the SOP procedure steps into runner steps. Each SOP procedure step
that calls a skill becomes a runner step with CONDENSED skill logic embedded.

For each skill execution step:
- State what data from Step 1 (or previous steps) to use
- Embed the skill's core output logic: tables, calculations, frameworks,
  specific formats
- Include the skill's quality checks
- Drop: frontmatter, license, metadata, "what makes this different",
  verbose context, role-setting preamble

For non-skill steps (manual checks, cross-references, SOP triggers):
- Translate directly as analysis/decision steps
- Include the specific thresholds and decision criteria from the SOP]

## Step [Final]: Assemble the [Output Name]

Combine all outputs into a single unified document:

[Output Document Title]

[Date or period reference]

[Section from Step 2]

[Content reference]

[Section from Step 3]

[Content reference]

[... all sections ...]

[Action items / Next steps]

[Prioritized actions from the analysis]

SOPs to Trigger


## Quality Check (Internal — never shown to the user)

Before presenting the output, verify:

| Check | Requirement |
|-------|------------|
| Complete | [Every input element appears in the output] |
| Honest | [Negative findings are surfaced, not hidden] |
| Specific | [Actions have exact language, not generic "follow up"] |
| Math correct | [Calculations verified against input numbers] |
| Cross-referenced | [Multiple skill outputs are reconciled] |
| Actionable | [Clear priorities for what to do next] |

Identify the weakest section. Rewrite it. Verify the rewrite before presenting.

## Rules

[Derive rules from three sources:
1. SOP Common Mistakes — convert each into a rule
2. Skill-specific rules — key constraints from each embedded skill
3. Output format rules — scanability, word limits, formatting]

---
Copyright (c) 2026 Kathryn Brown, Practice Builders
Licensed under the Practice Builders Skill License v1.0
See https://practicebuilders.ai/license for terms.

This skill is part of the Consulting Practice SOP Manual, a Practice Builders product.
Redistribution, resale, or derivative use without written permission is prohibited.

Generation Rules

  1. Condense skills to their core. Each full skill is ~1400 words. Condense to ~400-600 words per embedded skill. Keep: output structure (tables, formats), calculation logic, quality checks, decision thresholds. Drop: frontmatter, license, metadata, role-setting, "what makes this different", verbose context paragraphs, examples that don't drive the output.
  1. Single input collection. All data the runner needs — from prerequisites, skill inputs, and procedure step requirements — gets collected in Step 1. No scattered prompts.
  1. Unified output. The final step assembles one document. Not separate skill outputs. The document structure follows the SOP's expected outcome.
  1. Preserve specificity from skills. If a skill has specific thresholds (e.g., "stalled = 7+ days"), calculation formulas (e.g., "weighted pipeline: Lead 10%, Discovery 25%"), or format requirements (e.g., specific table columns), keep them. These are the value.
  1. Convert SOP cross-references. If the SOP says "trigger the Cold Outreach Batch SOP," include it as a checkbox in the output's "SOPs to Trigger" section.
  1. Common mistakes become rules. Each common mistake from the SOP translates into a rule at the bottom. Add skill-specific rules where they prevent bad output.
  1. Keep it scannable. This is an operating document. Short paragraphs, tables for structured data, bold for emphasis. No walls of text.
  1. Escape dollar signs. Use \$ for any dollar amounts to prevent LaTeX rendering in Notion.
  1. Platform-agnostic. No Claude-specific features. Works on Claude, ChatGPT, Gemini.
  1. kebab-case naming. The name: field uses lowercase kebab-case: weekly-pipeline-review-runner, case-study-capture-runner.