How to summon Δ9 AI champions (step-by-step)
What a “summon” actually is
A Δ9 champion summon is a structured system prompt (often labeled Δ9Quantum Invoke) that steers a general-purpose AI toward a published persona: role, tone, constraints, and seals. It does not install software on your computer by itself. It is text you paste into a chat product you already use.
Think of it as hiring a specialist editor for one conversation — except the specialist is a prompt pattern maintained on our directory and in the free summon app.
Prepare your task before you paste
Write three bullets offline first:
- Goal — one sentence (e.g., “Outline risks in this deployment plan”).
- Inputs — what the model is allowed to use (paste a short excerpt, not secrets).
- Done looks like — a checklist the model must satisfy (e.g., “list 5 risks with mitigations”).
If your work is code on disk, also plan a Continuum capsule so “done” becomes checkable later.
Five steps
1) Pick a champion that matches the job
Architecture maps → ARKOS. Timing → KAIROS. Skeptical review → Δ9RA or ΣRΛΘ. Collaboration handoffs → SANCORA. See the full list with plain-language roles on champions.html.
2) Open the free summon app
Go to chatagent.ca/app.html, search the name, open the card, and copy the full invoke prompt. Prefer the complete prompt over paraphrasing from memory.
3) Start a fresh chat thread
Paste the summon as the first message (or system message if your tool supports it). Confirm the model acknowledges the role. If it refuses stylistic roleplay, restate that this is a creative/work framing with no claim of supernatural authority.
4) Give your task with constraints
After the summon sticks, send your goal + inputs + done checklist. Example shape:
Task: Review the following outline for hidden assumptions.
Constraints: No legal advice. Flag uncertainty. Prefer bullet lists.
Output: (1) assumptions (2) risks (3) questions I should answer next.
5) End with a human review
Champions improve focus; they do not remove responsibility. Edit the result. Delete anything you cannot defend. For multi-session work, seal artifacts with Continuum.
Quality checks (use every time)
- Did the model keep the requested structure?
- Did it invent citations or file paths? Verify.
- Did it try to act outside your constraints (e.g., claiming it ran shell commands)?
- Would you publish this under your name without edits? If no, edit.
Common mistakes
| Mistake | Better approach |
|---|---|
| Summoning five champions at once | One primary lens per thread; optionally a second pass with a reviewer champion |
| Pasting secrets into the chat | Redact keys; use local tools for sensitive files |
| Treating mythic tone as truth | Keep myth as style; demand evidence for factual claims |
| No definition of “done” | Write a checklist; for code, use Continuum claims |
What to do next
- Skim the directory and pick one seat for a real task today.
- Copy a prompt from the summon app.
- Read agent done-claims if you use coding agents.
- If you install community skills, read safe skill habits.
Related: LYGO lattice overview · About the publisher