Features

Six engines.
One system.

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.

Feature 01

Hierarchical Guideline Resolver

Your process fixtures for AI. Set a master guideline at the root — the SOP that every prompt must follow. Teams add constraint layers for regions, products, or departments. When a prompt is requested, Proompt walks the tree and merges by specificity. Like poka-yoke on an assembly line — the prompt physically cannot be assembled without the right context.

  • Master → department → region → team layering
  • Automatic conflict resolution by specificity
  • Pin to a guideline version or always use latest
  • Audit trail of every resolution
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
proompt repo
# Versioning
$ proompt commit -m "update JP disclaimers for Q2"
v3.4 committed
 
# Branching
$ proompt branch "experiment/shorter-tone"
Branch created from main@v3.4
 
# Variants
$ proompt variant --of "social-post" --name "short-form"
Variant registered
 
# Tagging
$ proompt tag "social-post" --tags "marketing,japan,social"
3 tags applied
Feature 02

Prompt Repository

Every prompt is a traceable artifact with full audit trail. No more undocumented changes, no more "who changed the tone?" — every modification is versioned, branched, and attributable.

  • Versioning — every change tracked, rollback anytime
  • Branching — experiment without affecting production
  • Variants — multiple versions for different contexts
  • Tagging — organise by topic, team, use case
  • Fork & Star — share or lock down
  • Pin or latest — lock to a version or resolve from HEAD
Feature 03

Meta-Prompting Engine

One instruction in, one mistake-proof prompt out. Describe your need in plain language. Proompt generates a structured, guideline-aware prompt — like a jig that shapes raw material into a precision part. No prompt engineering expertise required.

  • Natural language → structured prompt output
  • Tool-aware formatting (Claude, GPT, Midjourney, Cursor)
  • Auto-references your resolved guidelines
  • Saves generated prompts to your repository
proompt meta
# Generate a prompt from plain description
$ proompt meta --desc "product launch email for Japan market"
--scope "brand/acme/japan" --tool "claude"
 
Resolving guidelines...
Generating structured prompt...
Prompt generated · 847 tokens · claude-optimised
Saved: prompts/japan-launch-email@v1
proompt enhance
# Enhance an existing rough prompt
$ proompt enhance --input "write a social post about our product"
--tool "claude" --scope "brand/acme"
 
Analysing prompt structure...
Adding role, context, constraints, format...
Injecting guideline context...
Enhanced · 23 tokens → 312 tokens
Feature 04

Prompt Enhancer

Have a rough prompt? Run it through the quality gate. Proompt adds guardrails — role definitions, context injection, constraints, tool-specific formatting — transforming a sketch into a specification.

  • Adds structure: role, context, constraints, format
  • Injects relevant guideline context automatically
  • Tool-specific formatting for better AI responses
  • Before/after diff view
Feature 05

Adherence Engine

The final inspection station before anything ships. Paste any AI-generated output and get a compliance score against your resolved guidelines. Like an end-of-line quality check — but for content, code, and copy.

  • Tone & voice alignment scoring
  • Approved terminology & prohibited language detection
  • Formatting & template rule compliance
  • Regional legal & policy validation
  • Code standard & architecture checks
  • Auto-fix suggestions with one-click remediation
🛡️ proompt adherence report87 / 100
Tone of voice
96%
Terminology
92%
Formatting
78%
Legal compliance
100%
Regional policy
60%
⚠ 2 violations found
→ JP regional disclaimer missing
→ Price format uses USD instead of JPY
↳ Auto-fix available
proompt run
# Full pipeline: resolve + execute + validate
$ proompt run --scope "brand/acme/japan"
--for "social-post" --tool "claude"
 
Guidelines resolved · 3 layers merged
Prompt assembled · 1,247 tokens
Executed via claude-sonnet-4
Adherence: 94/100 · 0 violations
Output delivered
Feature 06

Just Do It Mode

The full production line in one command. Resolve → assemble → execute → validate → deliver. Jidoka meets generative AI — if the output fails adherence, the line stops. Nothing ships with defects.

  • Full pipeline: resolve → assemble → execute → validate → deliver
  • Choose your AI: Claude, GPT, or any supported tool
  • Auto adherence check before delivery
  • Available via CLI, API, and web UI

Six engines.
One command.

Get early access and start managing prompts like a system, not a spreadsheet.