🌳
Guideline Resolver
Your guardrails, built into the line. Master guidelines cascade through team and regional layers — like poka-yoke fixtures that make it impossible to assemble a prompt without the right context.
Building in public — Early access coming soon
Poka-yoke
for AI prompts
Mistake-proof your AI workflows. A hierarchical prompt engine that bakes your guidelines into every prompt by design — so off-brand, off-spec, off-policy outputs become impossible, not just detectable.
# Resolve guidelines + generate + validate in one shot
$ proompt run --scope "brand/acme/japan" --for "social-post" --tool "claude"
→ Resolving guideline tree...
├─ master brand-guidelines-v3.2
├─ layer acme-corp/tone-of-voice
├─ layer acme-corp/japan/local-rules
└─ prompt social-post-generator
✓ 3 guidelines merged · executed via claude-sonnet
✓ Adherence: 94/100 · 0 violations
→ Output delivered.
Core Platform
Six engines. Zero defects.
Like poka-yoke on a production line — but for every prompt your team generates. Guidelines are built into the process, not bolted on after. Errors get caught at the source, not in the review.
📦
Prompt Repository
Version-controlled prompt libraries with branching, variants, and tagging. Every prompt is a traceable artifact — no more 'final_v3_REAL_final.txt' in a shared drive.
🧠
Meta-Prompting Engine
One instruction in, one mistake-proof prompt out. Describe what you need in plain language — Proompt generates a structured, guideline-aware prompt optimised for your target LLM.
⚡
Prompt Enhancer
Feed in a rough prompt. Get back one with guardrails — role definitions, context injection, constraints, and tool-specific formatting that prevents hallucination and drift.
🛡️
Adherence Engine
The final inspection station. Paste any AI output and get a compliance score against your resolved guidelines — tone, terminology, formatting, policy. Violations flagged, auto-fixes suggested. Nothing ships unchecked.
🚀
Just Do It Mode
The full production line in one command. Resolve guidelines → assemble prompt → call the AI → run adherence check → deliver validated output. Jidoka meets generative AI.
The Guideline Tree
Mistake-proof by design.
Not by review.
◆ acme-corp/MASTER
├─ brand-voice.md
├─ visual-identity.md
├─ engineering/EXTENDS
├─ code-standards.md
└─ api-conventions.md
├─ japan/EXTENDS
├─ local-tone.md
└─ marketing/EXTENDS
└─ social-post.promptRESOLVED
Build quality in.
Don't inspect it after.
1Set the standard — brand voice, coding conventions, design system. Your single source of truth, like an SOP on the production floor.
2Teams add fixtures — regional rules, product overrides, department conventions. They constrain the process, not restrict the people.
3Pull the andon cord — request a prompt or output. Proompt walks the guideline tree, assembles full context automatically.
4Ship with zero defects — master + layers + tool formatting + adherence check. Every output is inspected before it leaves the line.
Use Cases
One guideline tree. Zero prompt defects.
Every department that touches AI is a defect vector — or a quality gate. Proompt makes it the latter.
🌏
Marketing — Regional
Tokyo team writes a social post. Master brand voice + Japan tone rules + legal disclaimers — delivered market-ready.
💻
Engineering — Code
Engineer asks for a React component. Gets it in the team's exact code style with architecture principles inherited.
⚖️
Legal — Compliance
Paste AI-drafted content into adherence engine. Validates against compliance guidelines, flags issues, suggests fixes.
1
Solo developers &
prompt engineers
prompt engineers
2–50
Teams managing brand,
design & dev standards
design & dev standards
∞
Agencies managing
multiple client guidelines
multiple client guidelines
Stop inspecting.
Start preventing.
Mistake-proof your AI workflows. Join the waitlist for poka-yoke-grade prompt management.