Prompt Steering Workbench v0.2.0 Local · ready

One-shot intent → selective guidance → reusable prompt

Describe the outcome. Let the system assemble the work.

The workbench selects a validated workflow, applies known enterprise and editorial defaults, grills only the material gaps, and compiles an inspectable prompt specification.

This file runs locally. Saved sets remain in this browser and are not encrypted. Do not store credentials, secrets, tokens, or restricted source material.

Validated starters

Begin with a known workflow

Choose directly or describe the goal above.

Reusable configurations

Saved sets

Semantic configuration

Build the prompt specification

Current route

A stable path through the active specification.

No route

Canonical specification

No active request

Not selected

Model compiler

Render the same PromptSpec for the selected provider profile.

Output profile

Change the deliverable without replacing the underlying task recipe.

Selective clarification

Grill the remaining gaps

Mode

Smart asks only blocking or materially consequential questions.

Questions

No active gaps

0 remaining

Inspect before use

Review the compiled prompt

No compiled prompt

Complete the build first.

0 estimated tokens

Semantic discovery

Prompt library

0 results

Reusable configurations

Saved sets

Local storage notice. Saved configurations remain in this browser and are not encrypted by the application. Store configuration, not confidential source material.

Product contract

A prompt workbench, not a prompt wall.

The MVP tests whether one-shot semantic compilation plus selective clarification captures more material requirements with less effort than raw prompting.

What it does

Recommends a validated recipe or a composite saved set when the request spans task and output goals.

Applies visible, removable enterprise, editorial, SDLC, and SPA context with provenance.

Keeps inferred values separate from explicit answers and grills only unresolved blocking or material fields.

Compiles a provider-oriented prompt and a portable PromptSpec.

Saves reusable configurations locally.

What it does not do

It does not call a model or guarantee output quality.

It does not provide secure storage, authentication, or collaboration.

It does not treat a visually coherent route as proof of correctness.

It does not replace representative evaluations and human review.

Editorial lineage

Enterprise SDLC coverage

Research briefRFCADDADR setTechnical specImplementation planRunbookPostmortem

The Enterprise Architecture workflow can compile an editorial package spanning discussion, decision, design, delivery, operations, and learning. ADD is treated as Architecture Design Document and remains organization-renamable.

Release advancement

What changed in v0.2

Trust and control

Inherited defaults are now applied to the compiled prompt, scoped by relevance, and removable per build.

Request inferences remain distinct from user-provided answers and can be rejected.

Blocking fields can no longer be bypassed by turning Grill off.

Every PromptSpec includes a deterministic configuration fingerprint.

Faster composition

One-shot routing can synthesize a recipe and output profile, such as latest-standard research rendered as a searchable SPA.

Enhancement packs add coherent evidence, enterprise, editorial, or SPA release behavior.

Saved sets can be forked, pinned, renamed, exported, and reused.

The enterprise SDLC output profile links research, RFC, ADD, ADR, specification, delivery, operations, and learning artifacts.

Provenance

Artifact information

Version0.1.0
Generated2026-07-19
Knowledge foundationThe Communication Guide v2.2.0 and prompt-library research
DistributionSelf-contained HTML · offline-capable
StorageBrowser localStorage for versioned, forkable personal configurations

Reusable configuration

Save current set

Do not include credentials, tokens, secrets, or restricted source content.