| <!doctype html> |
| <html lang="de"> |
| <head> |
| <meta charset="utf-8"> |
| <meta name="viewport" content="width=device-width, initial-scale=1"> |
| <title>PromptBoost V2 Adapter</title> |
| <meta |
| name="description" |
| content="PromptBoost V2 operationalisiert einen User-Prompt als harness-neutralen, eval-faehigen Prompt-Entwurf." |
| > |
| <link rel="stylesheet" href="/src/styles.css"> |
| </head> |
| <body> |
| <header class="topbar"> |
| <a class="brand" href="#top" aria-label="PromptBoost V2 Start"> |
| <span class="brand-mark" aria-hidden="true">PB</span> |
| <span> |
| <strong>PromptBoost V2</strong> |
| <small>User Intent -> Harness Draft</small> |
| </span> |
| </a> |
| <nav class="topnav" aria-label="Hauptnavigation"> |
| <a href="#adapter">Adapter</a> |
| <a href="#score">Score</a> |
| <a href="#eval-lab">Eval Lab</a> |
| <a href="#export">Export</a> |
| <a href="#sources">Quellen</a> |
| </nav> |
| </header> |
|
|
| <main id="top"> |
| <section id="adapter" class="workspace" aria-labelledby="app-title"> |
| <div class="workspace-head"> |
| <div> |
| <p class="eyebrow">Phase 3 Adapter</p> |
| <h1 id="app-title">Aus User-Intent wird ein Harness-Prompt-Entwurf.</h1> |
| </div> |
| <p> |
| V2 trennt beobachteten User-Intent, Adapter-Scaffolding, blanke Harness Policy, |
| Model Adapter und Eval Plan. Der Adapter veraendert nicht die Aufgabe, sondern |
| operationalisiert sie als pruefbaren, portablen Entwurf. |
| </p> |
| </div> |
|
|
| <div class="app-grid"> |
| <section class="panel input-panel" aria-labelledby="input-title"> |
| <div class="panel-head"> |
| <div> |
| <p class="panel-kicker">Input</p> |
| <h2 id="input-title">User Prompt</h2> |
| </div> |
| <span class="badge stable">Intent bleibt stabil</span> |
| </div> |
|
|
| <textarea |
| id="rawPrompt" |
| spellcheck="false" |
| aria-label="Raw User Prompt" |
| >Fasse ein langes PDF quellentreu in Markdown zusammen und erstelle ein PDF-Render-Skript.</textarea> |
|
|
| <div class="control-grid"> |
| <label> |
| <span>Task Type</span> |
| <select id="taskType"></select> |
| </label> |
| <label> |
| <span>Model Adapter</span> |
| <select id="adapterProfile"></select> |
| </label> |
| <label> |
| <span>Harness Policy</span> |
| <select id="harnessPolicy"></select> |
| </label> |
| </div> |
|
|
| <div class="button-row"> |
| <button id="generateButton" class="primary-button" type="button">Boost Prompt</button> |
| <button id="copyPromptButton" type="button">Copy Prompt</button> |
| <button id="downloadJsonButton" type="button">Export JSON</button> |
| </div> |
| </section> |
|
|
| <section class="panel prompt-panel" aria-labelledby="prompt-title"> |
| <div class="panel-head"> |
| <div> |
| <p class="panel-kicker">Output</p> |
| <h2 id="prompt-title">Harness Prompt Draft</h2> |
| </div> |
| <span id="overallBadge" class="badge">Score</span> |
| </div> |
| <pre id="boostedPrompt" class="output-box" aria-label="Boosted Harness Prompt"></pre> |
| </section> |
|
|
| <section class="panel spec-panel" aria-labelledby="spec-title"> |
| <div class="panel-head"> |
| <div> |
| <p class="panel-kicker">Compiler View</p> |
| <h2 id="spec-title">Task Spec</h2> |
| </div> |
| <span id="wordBadge" class="badge">0 words</span> |
| </div> |
| <div id="taskSpecCards" class="spec-cards"></div> |
| </section> |
| </div> |
| </section> |
|
|
| <section id="score" class="section-band" aria-labelledby="score-title"> |
| <div class="section-head"> |
| <div> |
| <p class="eyebrow">Local Validation</p> |
| <h2 id="score-title">Der Boost bekommt nur einen Heuristik-Score.</h2> |
| </div> |
| <p> |
| Der lokale Score beweist keine Ueberlegenheit und misst keine Ergebnisqualitaet. |
| Er schaetzt nur Prompt-Shape-Signale wie Struktur, Eingabe-Adequacy, Trennung von |
| Intent und Scaffolding sowie sichtbare Verifikationshinweise. |
| </p> |
| </div> |
| <div id="scoreGrid" class="score-grid"></div> |
| </section> |
|
|
| <section id="eval-lab" class="section-band" aria-labelledby="eval-title"> |
| <div class="section-head"> |
| <div> |
| <p class="eyebrow">Eval Harness</p> |
| <h2 id="eval-title">Profilvergleich ueber Beispielprompts</h2> |
| </div> |
| <p> |
| Die Tabelle vergleicht lokal nur Prompt-Profile nach Heuristik-Score. Sie ist kein |
| Modellbenchmark; echte Modelllaeufe koennen spaeter ueber das PromptRunner-Interface |
| und einen JudgeRunner angeschlossen werden. |
| </p> |
| </div> |
| <div class="table-wrap"> |
| <table> |
| <thead> |
| <tr> |
| <th>Case</th> |
| <th>Task Type</th> |
| <th>Best Profile by Heuristic</th> |
| <th>Heuristic</th> |
| <th>Key Signal</th> |
| </tr> |
| </thead> |
| <tbody id="evalRows"></tbody> |
| </table> |
| </div> |
| </section> |
|
|
| <section id="export" class="section-band export-band" aria-labelledby="export-title"> |
| <div class="section-head"> |
| <div> |
| <p class="eyebrow">Portable Artifact</p> |
| <h2 id="export-title">Portables JSON-Artefakt fuer das groessere Agentur-Projekt</h2> |
| </div> |
| <p> |
| Dieses Objekt ist ein portables Export-Artefakt: beobachteter Task Spec, |
| Adapter-Scaffolding, Harness Policy, Adapterprofil, Prompt Draft, |
| Heuristik-Score und Eval Plan in einer transportierbaren Struktur. |
| </p> |
| </div> |
| <pre id="jsonExport" class="output-box json-box" aria-label="PromptBoost JSON Export"></pre> |
| </section> |
|
|
| <section id="sources" class="section-band" aria-labelledby="sources-title"> |
| <div class="section-head"> |
| <div> |
| <p class="eyebrow">Quellenrahmen</p> |
| <h2 id="sources-title">Dokumentation als Prior, Evals als Urteil</h2> |
| </div> |
| <p> |
| Die Adapter unterscheiden Doku-Priors von expliziten Adapter-Hypothesen. |
| Modelllaeufe muessen diese Hypothesen spaeter empirisch bestaetigen oder verwerfen. |
| </p> |
| </div> |
| <div class="table-wrap"> |
| <table> |
| <thead> |
| <tr> |
| <th>Familie</th> |
| <th>Prior / Quellenanker</th> |
| </tr> |
| </thead> |
| <tbody> |
| <tr> |
| <td>OpenAI / GPT 路 strong_prior</td> |
| <td><a href="https://developers.openai.com/api/docs/guides/prompt-engineering" rel="noreferrer">Prompt engineering</a> 路 <a href="https://developers.openai.com/api/docs/models/gpt-5.4" rel="noreferrer">GPT-5.4</a></td> |
| </tr> |
| <tr> |
| <td>Claude 路 strong_prior</td> |
| <td><a href="https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/claude-prompting-best-practices" rel="noreferrer">Prompting best practices</a> 路 <a href="https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/long-context-tips" rel="noreferrer">Long context</a></td> |
| </tr> |
| <tr> |
| <td>Gemini 路 strong_prior</td> |
| <td><a href="https://ai.google.dev/gemini-api/docs/prompting-strategies" rel="noreferrer">Prompt design strategies</a></td> |
| </tr> |
| <tr> |
| <td>Kimi 路 moderate_prior</td> |
| <td><a href="https://platform.moonshot.ai/docs/guide/prompt-best-practice" rel="noreferrer">Prompt best practices</a> 路 <a href="https://platform.moonshot.ai/docs/guide/kimi-k2-5-quickstart" rel="noreferrer">K2.5 quickstart</a></td> |
| </tr> |
| <tr> |
| <td>GLM / Qwen / DeepSeek / Mistral / MiniMax / Llama</td> |
| <td>Siehe Adapter-Evidenz im JSON Export; moderate_prior oder hypothesis bedeutet: Quelle reicht fuer Startannahmen, nicht fuer absolute Regeln.</td> |
| </tr> |
| </tbody> |
| </table> |
| </div> |
| </section> |
| </main> |
|
|
| <script type="module" src="/src/app.ts"></script> |
| </body> |
| </html> |
|
|
