CASE STUDY // FRAMEWORK

Prompt Workflow Library

A governed collection of reusable prompt workflows organized by job-to-be-done, with built-in risk controls and version history

SECTION // HOW THIS RELATES TO PROMPT VAULT
The Prompt Workflow Library and the Prompt Vault solve different problems for different users. This library is built for builders and maintainers: the people who design, govern, and evolve AI workflows. Every entry includes version history, risk levels, token intensity ratings, do-not-use-for boundaries, and the reasoning behind each design decision. It is the operational infrastructure behind the prompts. The Prompt Vault is built for end users: the people who just need a prompt that works right now without needing to understand why it works. It is the front-facing interface where governed, tested prompts are made accessible to a non-technical audience.

Think of it this way: this library is the kitchen. The Prompt Vault is the menu.

See the Prompt Vault
SECTION // PROBLEM
Most prompt collections are junk drawers. Someone saves a prompt that worked once, shares it in Slack, and three months later nobody knows if it still works, who maintains it, or whether it is appropriate to use for a given task. The result is inconsistent outputs, duplicated effort, and no institutional memory around what actually works. The gap is not prompts. It is the operational infrastructure around prompts that makes them repeatable, trustworthy, and safe to use at scale.
SECTION // WHAT I BUILT

A governed prompt workflow library organized by job-to-be-done rather than by tool or topic. Every workflow in the library is a structured operational asset, not a saved chat snippet.

EIGHT CATEGORIES IN THE LIBRARY

SEO Triage

Review audit export, classify issues, rank by impact.

Redirect Strategy

Upload 404 list, bucket URLs, recommend 301 targets.

Metadata

Input page title and article summary, generate title, meta description, and H1 options.

Stakeholder Comms

Convert technical issue into executive summary.

Jira Tickets

Convert finding into dev-ready ticket with acceptance criteria.

AI Search Discovery

Test query set, assess likely visibility gaps.

Portfolio Storytelling

Turn project into case study format.

Governance

Evaluate AI output risk before use.

EVERY WORKFLOW INCLUDES SEVEN FIELDS

PURPOSEWhat business problem this prompt solves.
BEST INPUTWhat the user should provide (CSV, URL list, draft copy, audit snippet, stakeholder goal).
PROMPT TEMPLATEThe reusable instruction.
OUTPUT FORMATTable, Jira ticket, executive summary, checklist, redirect map.
REVIEW LEVELLow, medium, or high risk.
EXAMPLE OUTPUTA sanitized example.
NOTES AND KNOWN PITFALLSWhere the model may overreach, assume too much, or need human review.

DO-NOT-USE-FOR

The "do-not-use-for" field is built into every workflow card. It sets explicit boundaries on where each prompt should not be applied, which is the guardrail that keeps well-intentioned prompting from creating downstream risk.

VERSION HISTORY

Version history is tracked on every workflow. When a prompt produces poor outputs, the failure pattern is logged and the prompt is refined. v1 produced too many clever titles. v2 constrained tone to professional, clear, and healthcare-adjacent language. That iteration log is institutional knowledge, not just a revision history.

SECTION // EXAMPLE IN PRACTICE

The following workflow card is drawn from the Professor Crawl SEO agent deployment at a national healthcare not-for-profit. It represents the full template in use against a real operational need.

404 Redirect Prioritization

PurposePrioritize broken URLs based on SEO value, user experience, and authority preservation.
Intended UserSEO specialist, web manager, content strategist.
Best Used WhenA site audit reveals 404 URLs, especially pages with backlinks, traffic, branded relevance, or internal links.
Inputs Needed404 URL list, backlink count if available, organic traffic or clicks if available, existing target URL candidates if available, content category or page type.
Prompt TemplateReview the following 404 URLs and recommend an action for each. Use these action categories: (1) 301 redirect to closest relevant live page, (2) Consolidate into stronger hub page, (3) Archive with no redirect if no SEO or user value, (4) Investigate before action. Prioritize URLs with backlinks, branded terms, organic traffic, or strategic content value. Return a table with: Source URL, Recommended action, Suggested target URL or target category, Priority, Rationale, Human review needed. Do not recommend 302 redirects unless explicitly requested.
Expected OutputA prioritized redirect and action table.
Review LevelHigh
Token IntensityMedium to Heavy
Quality ChecklistDoes each redirect preserve user intent? Are high-authority URLs prioritized? Are irrelevant redirects avoided? Are assumptions flagged? Are implementation-ready items clearly separated from investigation items?
Version Notesv1: Basic redirect recommendations. v2: Added archive and no-redirect option. v3: Added human review requirement and intent-preservation checks.

This agent is documented in full at /work/professor-crawl

SECTION // PROMPT ENGINEERING METHODOLOGY

The Prompt Workflow Library is built on three documented prompt engineering techniques that turn a collection of saved prompts into a governed, repeatable system.

Prompt Versioning

Every workflow card carries a version history. When a prompt produces poor outputs, the failure pattern is logged and the prompt is refined rather than kept or discarded wholesale. That iteration log is institutional memory: it records what changed, why it changed, and what the previous version did wrong. It is the difference between a prompt that drifts and a prompt that improves.

Constraint Boundary Design

The "do-not-use-for" field built into every card sets explicit boundaries on where a prompt should not be applied. Most downstream risk in well-intentioned prompting comes not from a bad prompt but from a good prompt used in the wrong context. Constraint boundary design makes that context visible on the card itself, so misuse is prevented at the point of use rather than diagnosed after the fact.

Flipped Interaction

The Best Input field on every workflow card is the documented, transferable version of a flipped interaction pattern. Instead of leaving the user to figure out what to provide, each card defines exactly what the AI needs before it can produce reliable output. The underlying prompt instructs the AI to ask for missing inputs rather than proceeding with incomplete information. This is the same pattern built conversationally into Judd: the AI leads the intake rather than waiting for a complete brief. In the library context, flipped interaction is operationalized as a field in a structured card rather than a wizard sequence. The principle is identical: the AI should gather what it needs before generating, not generate based on whatever it is given. The practical outcome is the same in both contexts: fewer confident-sounding wrong outputs, more usable first drafts, and a system that protects the user from their own incomplete prompting.

This pattern is also documented in the Judd case study, where it is implemented as a conversational wizard sequence rather than a structured card field.

SECTION // OUTCOMES
Turned a personal prompt collection into a governed operational system with defined scope, risk levels, and maintenance ownership. Established version control as a standard practice for prompt management, creating institutional memory around what works and why. Made prompt workflows teachable and transferable: any team member can pick up a workflow card and use it correctly without needing to ask the person who wrote it. Created explicit "do-not-use-for" boundaries that prevent well-intentioned misuse and reduce hallucination risk in high-stakes outputs.
SECTION // WHAT I'D DO NEXT
The library expands naturally as new workflows are validated in real use. The next priority categories are AI search discovery workflows (GEO, AEO, citation gap analysis) and stakeholder enablement workflows (business case builder, roadmap recommendation, risk translation). Longer term, the library becomes a shared team asset with contribution guidelines, a review process, and a deprecation policy for prompts that no longer perform.