lh-platform-ops

name	lh-platform-ops
description	Operate the LH Enterprise platform (LLM gateway console): query org usage/billing, cache-coverage audit, topup/refund, plans & members — via the lh-ops CLI.
version	0.1.0
platforms	["linux","macos","windows"]
metadata	{"hermes":{"tags":["lh","platform","ops","billing","llm-gateway","console"],"related_skills":[]}}

LH Platform Ops

Operate the LH Enterprise platform — an enterprise management console for an LLM gateway (new-api). This skill drives the platform through the lh-ops CLI (a platform-operator tool that talks to the LH backend's /platform/* API). Use it when the user asks to query or manage orgs, usage, billing, plans, members, or keys on LH.

Prerequisites

• The lh-ops CLI must be available on PATH. It's published to npm (npm i -g lh-ops) — ./setup.sh auto-installs it if missing. If lh-ops is missing or ~/.lh-ops/credentials.json is unset, run ./setup.sh in this skill's directory first — it installs lh-ops + captures the platform token + base URL.
• A scoped platform token (issued by an LH admin via lh-ops token issue). The operator scope set: read:audit,read:billing,read:keys,read:members,read:orgs,write:billing,write:orgs.

Commands (lh-ops)

Run lh-ops <command> --help for exact flags. Remote targets need --api-base <url> --token <tok> (or rely on ~/.lh-ops/credentials.json from setup).

Org & usage (read)

• lh-ops org list — list orgs (optional --status, --near-limit, --over-budget)
• lh-ops org usage --org <id> — all-time reconciliation: balance, deposits/charges, by-model, by-billing-decision
• lh-ops org statement --org <id> --month <YYYY-MM> — monthly statement
• lh-ops org tokens --org <id> — list the org's API keys

Billing (read + write)

• lh-ops cache-coverage — platform-wide cache-hit audit by model (which models return prompt-cache tokens + hit rate; flags 0 %-cache models with ⚠️)
• lh-ops pricebook list — price-book versions (which is ACTIVE)
• lh-ops topup --org <id> --amount <USD> — credit an org's balance (WRITE — confirm before running)

Plans / members / keys

• lh-ops plan show <planId> / plan update <planId> --file <dto> --if-match <ts>
• lh-ops member ... (incl. unfreeze, set-cap)
• lh-ops token issue --scopes <...> / token revoke <id> / token list — platform-token lifecycle

Catalog / audit

• lh-ops models list — upstream model catalog (canonical names)
• lh-ops audit query / audit list — audit log

Safety boundaries (IMPORTANT)

• READ commands (usage / cache-coverage / list / statement / models / audit): run freely, report results.
• WRITE commands (topup, pricebook publish/activate, plan update, token issue/revoke, org suspend): always state the exact target + amount/version and get explicit user confirmation before executing. These touch live billing or routing.
• NEVER attempt: server deploys, new-api channel/admin changes, DB mutations outside lh-ops, dangerous:* scopes, or anything not exposed via lh-ops. If asked, refuse and explain lh-ops is the sanctioned surface.
• Money is USD-canonical; the console shows a display currency via each org's contractRate. State amounts in USD unless the user asks for the display currency.

Auth notes

• Token is a platform-scoped JWT-like (lh_pat_...). Base URL is the LH backend (e.g. https://launchhill.com/api).
• Tokens are shown once at issue — capture immediately. If a token is lost, revoke + reissue.
• The model the agent itself uses is independent of this skill — point Hermes/OpenClaw at any provider (OpenRouter, etc.). This skill is about operating LH, not about using LH as a model.