NightWatch Skills — recipes for AI agents

NightWatch Live Intelligence is a real-time, multi-layer crypto asset intelligence system: it watches ~10 exchanges (global + Korean), grades every listed token's liquidity (A–F, 15-step detail), tracks per-venue deposit/withdrawal doors per chain, records cross-exchange spread history, and condenses all of it into two agent-facing gate indices — Coverage Grade ("how well do we actually see this asset") and BTI, the Bot-Tradability Index ("which bot strategy types are viable on it right now"). Everything is served fresh from the live pipeline; missing data is reported as missing, never faked.

One product, four layers

Layer	What it is	Where
MCP	Connection standard — 33 live tools, no auth to browse	`https://nightwatch-v1-api.onrender.com/mcp`
API	Raw data under a stable REST contract (free tier + Pro key `X-NW-User-Key` for deep fields)	`https://nightwatch-v1-api.onrender.com`
Skills	Task recipes (SKILL.md folders) — this repo	`skills/`
Trader	Open-source reference trading bot (paper-first, non-custodial)	jamboree777/nightwatch-trader

MCP endpoint

Code

https://nightwatch-v1-api.onrender.com/mcp

Streamable HTTP, 33 tools (token intel, microburst, warnings, mining, oracle bounties, Proof-of-Insight, SBT reputation, Quartermaster decision state, …). Works with Claude Code, Cursor, and any MCP client:

config.json

{
  "mcpServers": {
    "nightwatch": {
      "type": "http",
      "url": "https://nightwatch-v1-api.onrender.com/mcp"
    }
  }
}

Listed on the official MCP registry as io.github.jamboree777/nightwatch — metadata in server.json, update procedure in PUBLISH.md.

For LLMs / agents

llms.txt: https://nightwatch-v1-api.onrender.com/llms.txt — how to fetch one-URL markdown "wiki notes" per token (GET /kg/{TOKEN}.md).
Agent gateway page: https://nightwatch-v1-frontend.onrender.com/agent

Skills catalog

Each folder under skills/ is one recipe: a SKILL.md any agent can load and follow. All endpoints referenced are live and were verified returning 200 at publish time.

Skill	What it does
`token-intel-check`	Read the three gates (Coverage / Rating / BTI) from one `/intel` call before touching a token
`arb-route-check`	Check transfer doors + spread history before any cross-exchange move
`coverage-screen`	Screen a token list down to bot-tradable candidates
`kimchi-premium-read`	Read the Korea-axis premium correctly (FX vs stablecoin basis)
`contribute-facts`	WORK-Claw: submit sourced facts to the Knowledge Graph and earn Cherry
`fund-operating-procedure`	The fund's invariant arbitrage pipeline (spread → doors → identity → grade → costs → repeat caps → settle) with a live endpoint per step
`capital-allocation-quartermaster`	Quartermaster: two-book capital segregation, attribution-derived venue floors, rebalance-through-arbitrage, locked collateral pools

Honesty rules baked into every skill: free vs Pro-key fields are labeled; unknown never means no; no capability is claimed that the live API does not actually serve.

License

Documentation and skill recipes in this repository: CC BY-NC-SA 4.0 — see LICENSE.md. The reference trading bot is separately MIT-licensed in its own repo.