The substrate the constellation reads first, every time.
Built via a structured intent-archaeology interview. Twenty questions, plus three writing samples, plus your three favorite arguments. Output is a compressed profile under two thousand tokens that the constellation loads at the top of every session. The model stops guessing who you are because it already knows.
Bans one hundred plus AI tells (delve, unlock, leverage, tapestry, navigate, harness, the list keeps growing). Kills negative parallelism, the it's-not-X-it's-Y construction that every model defaults to. Caps paragraphs at three sentences. No em-dashes. No en-dashes. Spaced hyphens only. Every draft passes through the style-guard skill before it ships.
Six to eight questions. Under one thousand tokens. Updated quarterly. What you are shipping this quarter. What number it has to hit. What you turned down to focus on it. The constellation reads this to know when to push back on a request that drifts from the mission.
Five engine variants, same contract. Thirty principles. Five core rules. Triple-I integrity gate. A binding rule that done means customer-usable. Claude gets the structured XML version. ChatGPT, Gemini, Grok, and Perplexity each get the same contract in their native idiom. Version bumps move all five engines together. No drift.
Runs before every send. Strips slop. Catches prompt injection. Re-rewrites messy prompts with action verbs and length caps. Flags depth mismatch when the answer is heavier than the question deserves. The skill is packaged, versioned, and shipped with every tier. You do not have to remember to run it.
All five files. All five engines. All five skills. Pre-loaded.
Start at app.dreameros.app See the full vault