Kernel Skill — `zeq-os-mi-kernel`

The Zeq kernel as a SKILL.md any LLM client can read. Source of truth lives at one URL: change it once, every install re-syncs on next pull.

A skill is a markdown file with YAML frontmatter that LLM clients load as a system prompt. The Zeq kernel skill is the boot sequence for any agent operating the Zeq engine — it pins the 1.287 Hz HulyaPulse, mandates KO42 on every call, and wires the 7-step protocol so every output carries the CKO envelope and zeqProof.

Three canonical endpoints, one source of truth:

Endpoint	Format	When to use
`GET /api/kernel/prompt`	`text/plain`	Raw boot sequence — paste straight into a system prompt field
`GET /api/kernel/skill.md`	`text/markdown`	Full SKILL.md with YAML frontmatter — drop into a skills directory
`GET /api/kernel/skill`	`application/json`	Manifest with version, sha256, per-platform install steps

The markdown is generated from the same buildKernelMarkdown() builder that serves /.well-known/zeq-os-kernel.md, so no two surfaces ever drift.

Install

Claude Code

mkdir -p ~/.claude/skills/zeq-os-mi-kernel
curl -sL https://zeqapi.com/api/kernel/skill.md \
  -o ~/.claude/skills/zeq-os-mi-kernel/SKILL.md
# Restart Claude Code or reload skills

The skill activates automatically when a user mentions Zeq, requests physics computation, or pastes a Zeq operator ID.

Cowork

In a fresh plugin, paste the markdown into the plugin's SKILL.md:

curl -sL https://zeqapi.com/api/kernel/skill.md

Cursor

mkdir -p .cursor/rules
curl -sL https://zeqapi.com/api/kernel/prompt \
  -o .cursor/rules/zeq-kernel.md

OpenAI Assistants

Use the prompt as the Assistant's system prompt:

curl -sL https://zeqapi.com/api/kernel/prompt
# Paste the response into the "Instructions" field on platform.openai.com

Manual install

Open https://zeqapi.com/api/kernel/skill.md in a browser, save the page, drop it anywhere your LLM reads system prompts from.

Verify integrity

The JSON manifest carries a SHA-256 of the markdown bytes:

curl -sL https://zeqapi.com/api/kernel/skill | jq .sha256
# Expected:  9fe17bfd7c4302b8…

Recompute locally to confirm:

curl -sL https://zeqapi.com/api/kernel/skill.md | shasum -a 256

The two hashes must match. If they don't, you're on a stale CDN edge or someone is intercepting the file.

What the skill enforces

Every call your agent makes through the Zeq engine inherits these rules from the SKILL.md:

Rule	What it means
KO42 on every call	`operators[0]` is always `KO42` — the metric tensioner that phase-locks output to 1.287 Hz
Cross-domain experiments	Each compute combines operators from ≥2 physics domains (quantum + gravitational, etc.)
Real inputs only	Stellar masses, atomic radii, particle energies — no `m=1, v=1` toys
Span the scale range	Across 20 experiments, cover quantum → cosmological
Side-by-side comparison	Every result reports the textbook formula + value AND the Zeq result with R(t), phase, zeqond, master_sum, equations[], zeqProof
Variance gate	If zeqond variance > 0.0091, the kernel triggers a metric shift before returning

Without the skill installed, an LLM connecting to /api/zeq/compute will still get valid results, but will fall back to its own (often weaker) framing for what to compute. The skill is what turns the API from "a calculator" into "an instrument."

What's in the manifest

{
  "ok": true,
  "schema": "zeq.skill.v1",
  "name": "zeq-os-mi-kernel",
  "version": "1.287.5",
  "license": "CC BY 4.0",
  "homepage": "https://zeqapi.com/sdk",
  "sources": {
    "kernel_json":    "https://zeqapi.com/api/kernel",
    "prompt_text":    "https://zeqapi.com/api/kernel/prompt",
    "skill_markdown": "https://zeqapi.com/api/kernel/skill.md",
    "well_known":     "https://zeqapi.com/.well-known/zeq-os-kernel.md",
    "live_pulse":     "https://zeqapi.com/api/zeq/pulse"
  },
  "bytes": 8517,
  "sha256": "9fe17bfd7c4302b8…",
  "install": { "claude_code": { … }, "cowork": { … }, "cursor": { … } }
}

Refreshing the skill

Re-fetch any time you want to pick up kernel updates. Most clients pull-on-load, so a restart is enough:

curl -sL https://zeqapi.com/api/kernel/skill.md \
  -o ~/.claude/skills/zeq-os-mi-kernel/SKILL.md

Add this to a cron / launchd job if you want auto-pull on a schedule:

# Every 7 days, refresh the kernel skill
0 4 * * 0 curl -sfL https://zeqapi.com/api/kernel/skill.md \
  -o ~/.claude/skills/zeq-os-mi-kernel/SKILL.md

Federation

When other Zeq domains come online (zeqond.com, hulyapulse.com, …) each ships its own /api/kernel/skill.md. You can install multiple side-by-side — the skills don't conflict because each is namespaced under its own slug:

# zeq.dev kernel
curl -sL https://zeqapi.com/api/kernel/skill.md \
  -o ~/.claude/skills/zeq-os-mi-kernel/SKILL.md

# zeqond.com kernel (when live)
curl -sL https://www.zeqond.com/api/kernel/skill.md \
  -o ~/.claude/skills/zeqond-kernel/SKILL.md

Each domain pins its own pulse rate / operator emphasis but agrees on the bridge t_Zeq = t_Unix / 0.777.

Pair with MCP

The SKILL.md tells the LLM what to do. The MCP server gives it the tools to do it. Install both for the full loop:

Skill → loaded as system prompt → mandates KO42, 7-step, side-by-side comparison
MCP → exposes zeq_compute, zeq_pulse, zeq_verify, etc. as callable tools

{
  "mcpServers": {
    "zeq": {
      "command": "npx",
      "args": ["-y", "zeqsdk-mcp"],
      "env": {
        "ZEQ_API_URL": "https://zeqapi.com",
        "ZEQ_API_KEY": "zeq_ak_…"
      }
    }
  }
}

With the skill installed and the MCP server connected, your LLM can run a 20-experiment cross-domain batch end-to-end without any human prompting.

Install​

Claude Code​

Cowork​

Cursor​

OpenAI Assistants​

Manual install​

Verify integrity​

What the skill enforces​

What's in the manifest​

Refreshing the skill​

Federation​

Pair with MCP​

See also​