LLM Skill
An LLM skill that audits your product design artifacts, checks referential integrity between them, and coaches you through the Product Cycle — from personas through analytics. Works with Claude Code, OpenClaw, and other LLM coding tools that support custom commands.
The skill audits your project's product design artifacts in cycle order, checking that each stage is present and that artifacts reference each other correctly. It coaches rather than gatekeeps — reporting what's missing and suggesting what to do next.
Checks that personas exist with roles, responsibilities, goals, and context. Flags personas with no user stories.
Validates story format (As a / I want / So that), checks for acceptance criteria in Given/When/Then format. Flags orphaned stories with no persona.
Extracts domain nouns from stories into a table. Flags inconsistent terminology and concepts that appear in stories but aren't inventoried.
Checks for a DATA_MODEL.md with a mermaid ER diagram. Validates that every concept has a corresponding entity with defined relationships.
Checks that the backlog is current, stories are prioritized and refined with acceptance criteria and clear scope.
Checks for instrumentation that closes the feedback loop — analytics tied back to personas and use cases, informing new stories.
The skill's primary job is checking that artifacts reference each other correctly. These links are the skeleton of the framework — break one and the cycle breaks.
Every story names the persona it serves. Every persona has at least one story.
Stories introduce domain nouns. Every concept traces back to the stories where it appears.
Every concept becomes a data model entity (or is explicitly excluded with rationale).
Stories with acceptance criteria become backlog items. Items without criteria are unrefined.
Instrumentation measures what users do. Analytics events map to personas and use cases.
Observed behavior updates market understanding, validates persona assumptions, generates new stories.
Here's how one feature traces through every artifact, with each stage referencing the ones before it.
form_manager — manages forms for customer feedback
As a form_manager, I want to export form responses as CSV, so that I can analyze feedback in a spreadsheet. Acceptance Criteria: - Given a form with at least one response, When I click "Export CSV", Then a CSV file downloads containing all responses with headers matching question labels.
Form has_many Responses Form has_many Questions Response belongs_to Form
Export CSV — form_manager — high priority — acceptance criteria defined
Each artifact references the ones before it. Break any link and you have a referential integrity violation.
When you run /product-cycle,
you get a structured report:
## Product Cycle Audit
### Summary
Personas: 2 Stories: 5 Concepts: 8
Data Model: exists Backlog: exists Analytics: missing
### Cycle Integrity
✓ Personas ✓ Stories ✓ Concepts ✓ Data Model ✓ Backlog ✗ Analytics
Violations: 1 orphaned story, 2 concepts not in data model
### Findings
- Personas: ✓ Both personas have stories and acceptance criteria
- Stories: ⚠ "Export Data" has no persona (orphaned)
- Concepts: ⚠ "notification" and "template" appear in stories
but are missing from CONCEPT_INVENTORY.md
- Data Model: ⚠ 2 concepts have no corresponding entity
- Backlog: ✓ 4 of 5 stories refined with acceptance criteria
- Analytics: ✗ No instrumentation found
### Recommended Next Actions
1. Add analytics instrumentation to close the feedback loop
2. Add "notification" and "template" to CONCEPT_INVENTORY.md
3. Add corresponding entities to DATA_MODEL.md
4. Assign orphaned story "Export Data" to a personaThe skill is a single markdown file. It works with any LLM coding tool that supports custom commands — Claude Code, OpenClaw, and others. Install it globally or per-project.
Available in every project you open with Claude Code:
mkdir -p ~/.claude/commands
curl -o ~/.claude/commands/product-cycle.md \
https://raw.githubusercontent.com/afomi/\
product-cycle-skill/main/product-cycle.mdThen run:
/product-cycleCommit the file so everyone on the team gets the skill:
mkdir -p .claude/commands
curl -o .claude/commands/product-cycle.md \
https://raw.githubusercontent.com/afomi/\
product-cycle-skill/main/product-cycle.md
git add .claude/commands/product-cycle.md
git commit -m "add product-cycle skill"
The skill looks for artifacts in the project root or a
product-cycle/ directory:
| Artifact | File patterns |
|---|---|
| Personas | PERSONAS.md, personas/*.json, personas/*.md |
| User Stories | USE_CASES.md, use_cases/*.json, stories/*.md |
| Concept Inventory | CONCEPT_INVENTORY.md, CONCEPT_INVENTORY.csv |
| Data Model | DATA_MODEL.md (mermaid erDiagram) |
| Backlog | BACKLOG.md, GitHub Issues |
| Analytics | ANALYTICS.md, instrumentation in code |
Install the skill, open any project, and run
/product-cycle.