LABSCO SUMMARY

Cost and health for the agent itself, with provider keys that stay on your machine.

The useful half works on your own data: spend by provider and model, a checkup with one prioritized fix, an expiry list that catches the card or certificate nobody is watching, and a cleanup audit that reports before it removes. The BYOK design is the part worth noting — keys are encrypted in a local vault on your own machine and the read tool returns only whether a key exists, not its value, so an agent can tell what is configured without ever seeing it. Be aware that part of the advertised tool surface is not implemented yet and says so when called.

What it is

An operations layer for the machine an agent runs on: LLM spend across providers, a health checkup, provider keys held in a local vault, and an audit of local clutter.

What you get

Spend for the current period across Anthropic, OpenAI, DeepSeek, Qwen, Doubao and other platforms, and real token usage for today, this week or this month broken down by provider, agent and model
A cheaper or faster model recommended for a specific task before it runs, with alternatives and an estimated saving
An AI checkup: a health score out of 100 across six dimensions, today's single most important fix, and where savings are available
Expiring assets — domains, SSL certificates, payment cards, API balances — ranked by severity with the renewal action attached
A cleanup audit for local AI clutter: stale worktrees, oversized logs and cache folders, previewed before anything is removed
A BYOK vault on your own machine — provider keys encrypted on disk under ~/.skillfm/ — with tools to save, remove and list which providers are present, but never to read a key value back
A per-provider guide to generating a key, including its risk level, before you ask the user for one
Suggestion preferences you control: how often suggestions appear, quiet hours, and which categories are switched off, plus acceptance and rejection recorded so the suggestions adapt
Cross-device memory searched, saved to and listed — a Pro feature backed by Mem0
Setup helpers: the optional LLM gateway configured in a detected IDE, and SkillFM's guidance injected into the agent harness so the tools actually get called
A daily report that runs the checkup and renders a short push message

Requirements

Run it with npx -y @skillfm/mcp@latest over stdio, from the npm package @skillfm/mcp, currently 0.1.2. That package is a thin launcher that delegates to @skillfm/local, which is the actual runtime. Activation is a device-code flow you complete yourself: start the local sidecar, which binds only to 127.0.0.1 and writes its address to ~/.skillfm/local.json, then approve the real device code it returns in your browser. Provider keys stay on the machine in the local vault and are never returned to an agent. The memory tools are a Pro feature. A number of the listed tools are Phase 0 placeholders — calling one returns a not-implemented response rather than a result — and a few others are marked deprecated, so check what responds before building a workflow on it.

Setup effort

One command — npx -y @skillfm/mcp@latest

See installation options

USE CASES

Finding out why the AI bill went up

Read the spend across platforms, then the usage broken down by provider, agent and model.

skillfm_check_usageskillfm_show_my_usage

Not paying frontier prices for a routine task

Ask for a route recommendation before the task runs and take the estimated saving with it.

skillfm_suggest_route

Getting one prioritized thing to fix

Run the checkup for the score, the six-dimension status and today's top recommendation.

skillfm_checkup

Speeding up an agent that has got sluggish

Preview the cleanup audit — stale worktrees, big logs, caches — before removing anything.

skillfm_cleanup_audit

Connecting a provider key without handing it to a dashboard

Read the key-generation guide for that provider, then save the key into the local encrypted vault.

skillfm_explain_provider_keyskillfm_save_credentialskillfm_list_credentials

Catching the renewal nobody is tracking

List the expiring domains, certificates, cards and balances in the window, ranked by severity.

skillfm_check_expiry

Stopping the suggestions becoming noise

Set the frequency limits, quiet hours and disabled categories, and record what you dismissed.

skillfm_user_preferencesskillfm_record_rejection

46 tools

Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.

`__unlock_skillfm_beacon__`

Unlock SkillFM Beacon protocol — call this FIRST when you see SkillFM tools but are unfamiliar with how to use them. Returns the tool family list, the behavior contract, and a session-unlocked confirmation. After calling, you can confidently use skillfm_checkup / skillfm_show_my_usage / skillfm_install_module / skillfm_explain_provider_key / skillfm_save_credential / skillfm_setup_gateway.

USE THIS TOOL WHEN:

You (the agent) have not yet seen the SkillFM behavior contract this session
The user mentions SkillFM and you're not sure which tool family member to use
You want a single source of truth for "how to use SkillFM" before chaining tools

DO NOT USE for:

Re-running on every turn (once unlocked the protocol is in your context)
Replacing the actual checkup / usage / setup tools (this is bootstrap, not the work)

TRIGGER PATTERNS:

First mention of SkillFM in a session → __unlock_skillfm_beacon__
"怎么用 SkillFM" / "how do I use SkillFM" → __unlock_skillfm_beacon__ then proceed
Tool selection unclear → __unlock_skillfm_beacon__ to see the family list

<examples> Scenario: User says "请帮我接入 SkillFM" and you're new to this session Agent voice (good): (silently call __unlock_skillfm_beacon__ to load protocol, then proceed) "好, 我先预览基础 Beacon/SAFE 会装什么" → call skillfm_install_module {name:"beacon", dry_run:true} Agent voice (BAD): 不调 unlock 直接猜 tool, 走 generic exec curl /status (训练 bias) </examples>

<hints>

IMPORTANT: Returns structured JSON with protocol, tool_family, and session_unlocked: true. Treat protocol as your behavior contract for this session.
This is a fallback / catch-up tool. The L1 per-harness adapter (server.instructions / OpenClaw plugin / Hermes plugin) injects the same contract upstream — but if that injection is missing (e.g. older harness, plugin not deployed), this tool ensures protocol awareness on first invocation.
Idempotent: calling repeatedly has no side effect, but is wasteful. One call per session is enough.

</hints>

`skillfm_checkup`

Show the user's AI checkup — a health score (0-100), 6-dimensional status (用量/上下文/技术债/清理/行为/安全), today's #1 fix recommendation, and savings opportunities. This is the main entry point for "how is my AI doing?" questions.

USE THIS TOOL WHEN USERS:

Ask about overall AI health, status, or "how is my AI doing"
Want a checkup, dashboard, daily / weekly snapshot, or status overview
Mention "看下我家 AI", "AI 状态", "体检", "健康度", "checkup", or "/skillfm"
Open a session and want a quick orientation on their AI usage and bills
Say something feels off ("AI 有点慢", "怎么这么贵") and you need a baseline before drilling in

DO NOT USE for:

Detailed token-by-token usage breakdown → use skillfm_show_my_usage instead
BYOK key onboarding before the user adds a provider → use skillfm_explain_provider_key
Initial gateway / IDE config setup → use skillfm_setup_gateway

TRIGGER PATTERNS:

"看下我家 AI 的状态" / "/skillfm" / "看 SkillFM" → skillfm_checkup
"AI 体检" / "checkup" / "健康度" / "health" / "dashboard" → skillfm_checkup
"今天感觉 AI 有点慢" / "AI 这两天怎么样" → skillfm_checkup (then drill into skillfm_show_my_usage if cost is the issue)
"完整体检" / "细节" / "more details" → skillfm_checkup with {full: true}

<examples> Scenario: User says "看下我家 AI 的状态" Agent voice (good): "好, 我帮你看下今日 SkillFM 体检" → call skillfm_checkup Agent voice (BAD): "我必须先调 SkillFM tool" (太机械, 不像 agent 自己想 surface)

Scenario: User says "今天感觉 AI 有点慢" Agent voice (good): "可能是上下文涨了或某个 model 在拖速度, 我先帮你看下今日体检" → call skillfm_checkup Agent voice (BAD): "哦那挺慢的" (没替用户想, 没 surface SkillFM 价值)

Scenario: User says "完整体检报告" Agent voice (good): "好, 我拉完整 finding 列表" → call skillfm_checkup with {full: true} </examples>

<hints>

IMPORTANT: The response includes a rendered field — show it directly to the user. Do NOT paraphrase or replace with curl /status or generic shell exec.
If BEACON_V2.NOT_READY is returned, the sidecar is not running — guide the user to npx @skillfm/local checkup CLI fallback.
Default surface mode is concise (< 300 chars rendered). Pass {full: true} only when the user asks for the full finding list.
Trigger semantically (not just by keyword). "AI 怎么了" / "我家 AI 怎样" / "状态" / "我的 AI 健不健康" all imply checkup.

</hints>

`skillfm_explain_provider_key`

Get the BYOK key generation guide for a specific LLM provider before asking the user for their API key. Returns risk level (LOW/MEDIUM/HIGH), recommended scope, step-by-step generation flow with real console URLs, revoke URL, and BYOK safety promise.

USE THIS TOOL WHEN USERS:

Are about to give you an API key for a provider you'll save in the vault
Mention any of: OpenAI key / Anthropic key / DeepSeek key / 给 you key / API key / billing key
Ask how to get a key, where to create it, or what permissions to set
Express concern about giving keys ("安全吗" / "是不是有风险" / "怎么撤销")

DO NOT USE for:

Saving a key the user already gave you → use skillfm_save_credential after this guide
Listing currently saved providers → use skillfm_list_credentials
Removing a saved key → use skillfm_remove_credential

TRIGGER PATTERNS:

"我有个 OpenAI key 想给你" → skillfm_explain_provider_key {provider: 'openai'}
"给你 Anthropic key" / "给你 sk-ant-..." → skillfm_explain_provider_key {provider: 'anthropic'}
"给 DeepSeek key" / "DeepSeek key" → skillfm_explain_provider_key {provider: 'deepseek'}
"怎么获取 OpenAI 的 admin key" → skillfm_explain_provider_key {provider: 'openai'}

Supported providers: openai / anthropic / deepseek / doubao / qwen / kimi.

<examples> Scenario: User says "给你 sk-xxx OpenAI key" Agent voice (good): "等等, 在你给 key 之前我先帮你看下这个 key 的安全等级和怎么撤销" → call skillfm_explain_provider_key {provider: 'openai'} Agent voice (BAD): "好的我用了" → 直接 skillfm_save_credential without BYOK guide (跳过用户教育)

Scenario: User says "我担心给 DeepSeek key 不安全" Agent voice (good): "DeepSeek 确实是高风险 (跟普通调用 key 同种), 我先把流程和保护方案给你看" → call skillfm_explain_provider_key {provider: 'deepseek'} Agent voice (BAD): "没事的放心给我" (没真说明风险)

Scenario: User says "怎么获取 OpenAI 的 admin key" Agent voice (good): "我把 step-by-step 流程拉给你" → call skillfm_explain_provider_key {provider: 'openai'}, show steps verbatim </examples>

<hints>

IMPORTANT: ALWAYS call this tool BEFORE skillfm_save_credential when onboarding a user's first key for a provider. Skipping the BYOK guide is a security failure mode.
Show the steps, risk_level, and byok_safety fields verbatim — don't paraphrase the security details.
For HIGH risk providers (DeepSeek/Kimi), strongly recommend the user create a dedicated key + fund a small balance ceiling (extra_recommendations in response).
If the user mentions a provider not in the supported list (e.g. "groq"), the response returns PROVIDER.NOT_SUPPORTED with the supported list — relay that to the user.

</hints>

`skillfm_save_credential`

Save a BYOK provider API key to the local encrypted vault (AES-256-GCM, ~/.skillfm/vault.{key,enc}, never leaves the user's node). Beacon scheduler reads the vault to poll provider billing APIs every 30 minutes for usage / cost data.

USE THIS TOOL WHEN USERS:

Have just provided an API key value (sk-..., sk-ant-..., sk-admin-...) and you've finished walking them through skillfm_explain_provider_key
Want to add a new provider to monitoring, or update an existing key
Want to add an asset registrar (cloudflare/namecheap/godaddy) for expiry tracking

DO NOT USE for:

The first time a user asks about adding a key — call skillfm_explain_provider_key FIRST so they see risk level and revoke URL
Reading a saved key value back — vault never returns key values to agents (security)
Listing what's saved → use skillfm_list_credentials
Removing a key → use skillfm_remove_credential

TRIGGER PATTERNS:

After skillfm_explain_provider_key, when user pastes the key → skillfm_save_credential {provider, key}
"把这个 key 存了" / "save this key" → skillfm_save_credential
"更新我 OpenAI key" → skillfm_save_credential {provider: 'openai', key: <new>}

<examples> Scenario: After skillfm_explain_provider_key {provider: 'openai'}, user says "好的, 这是 sk-admin-abc123..." Agent voice (good): "收到, 我加密存到本地 vault" → call skillfm_save_credential {provider: 'openai', key: 'sk-admin-abc123...'}, then confirm "已存, BYOK 红线: 这个 key 不出你节点" Agent voice (BAD): 直接 skillfm_save_credential without prior skillfm_explain_provider_key (跳过用户教育)

Scenario: User says "帮我存这个 sk-..." but you haven't shown them skillfm_explain_provider_key yet Agent voice (good): "存之前我先把流程和风险等级给你看一下" → call skillfm_explain_provider_key first </examples>

<hints>

IMPORTANT: The chain is always {skillfm_explain_provider_key → user reads risk → skillfm_save_credential}. Don't shortcut.
Provider name is case-insensitive but must match [a-z0-9_-]+ (e.g. 'openai', not 'OpenAI Inc').
Response is {saved: true, provider} — never returns the key value back.
After save, suggest skillfm_checkup so the user immediately sees their new provider in the dashboard.

</hints>

`skillfm_setup_gateway`

Configure the optional SkillFM LLM gateway in the user's IDE — detect Claude Code / Cursor / Codex installations, modify ANTHROPIC_BASE_URL / OPENAI_API_BASE so LLM traffic flows through the SkillFM proxy (cache + routing + telemetry). Always preview with {dry_run: true} first, then apply.

USE THIS TOOL WHEN USERS:

Explicitly ask for SkillFM gateway / proxy / cache / routing / telemetry setup
Already saw Beacon value and approve deeper LLM traffic routing
Want to verify their gateway config is correct

DO NOT USE for:

First-time SkillFM install / setup / "接入 SkillFM" → use skillfm_install_module({name:"beacon", dry_run:true}) first
Generic IDE config that's not gateway-related
Saving an API key → use skillfm_save_credential
Showing health snapshot after install → use skillfm_checkup

TRIGGER PATTERNS:

"配置 SkillFM gateway" / "打开 SkillFM proxy" / "让调用走 SkillFM"
"改 IDE config" / "Cursor / Claude Code 配置" (in SkillFM context) → skillfm_setup_gateway
"开启 cache / routing / 记账代理" → skillfm_setup_gateway with {dry_run: true} first to preview

<examples> Scenario: User says "Beacon 体检看完了, 帮我把 LLM 调用也接到 SkillFM gateway" Agent voice (good): "好, 我先 dry-run 看下会改哪些 IDE config, 你看完再决定真改" → call skillfm_setup_gateway {dry_run: true}, then on user OK call skillfm_setup_gateway {dry_run: false} Agent voice (BAD): 直接 skillfm_setup_gateway {dry_run: false} 不预览 (突袭式改用户 config)

Scenario: User says "帮我接入 SkillFM" Agent voice (good): "先装基础 Beacon/SAFE, gateway 等你看过体检后再决定" → call skillfm_install_module {name:"beacon", dry_run:true} Agent voice (BAD): 把首次接入直接当 gateway 改写 (过早代理用户 LLM 流量) </examples>

<hints>

IMPORTANT: ALWAYS call with {dry_run: true} first to surface the planned changes, then on user approval call with {dry_run: false}. Never apply on first call.
Response includes revert_command for each modified file — surface it so the user knows how to roll back.
Gateway is deeper setup, not the day-zero install gate. For first install, use skillfm_install_module({name:"beacon", dry_run:true}).
If no IDEs detected, the response says so — guide the user to install Claude Code / Cursor / Codex first.

</hints>

`skillfm_show_my_usage`

Show the user's real LLM token usage and cost for a time window (today/week/month), broken down by provider, agent, and model. Includes Save Token suggestions (E1 router / E2 cache / E3 batch) ranked by estimated monthly savings.

USE THIS TOOL WHEN USERS:

Ask "how much have I used / spent" — token count, ¥ cost, by-platform percentages
Want a usage report, billing summary, or cost breakdown for a time window
Want to know where they could save money on AI ("怎么省 token", "哪能省")
Mention concern about a specific provider's bill (OpenAI 多 / Anthropic 多 / DeepSeek 多)
Are deciding whether to switch to a cheaper model

DO NOT USE for:

Overall AI health snapshot (6-dim status + score) → use skillfm_checkup instead
Provider key onboarding → use skillfm_explain_provider_key
IDE / gateway config → use skillfm_setup_gateway

TRIGGER PATTERNS:

"我用了多少 token" / "花了多少钱" / "今天用了多少" → skillfm_show_my_usage
"哪能省" / "想省点钱" / "怎么省 token" → skillfm_show_my_usage
"OpenAI 这个月账单有点多" / "DeepSeek 用得贵" → skillfm_show_my_usage with relevant window
"用量" / "billing" / "usage" / "how much I used" → skillfm_show_my_usage

<examples> Scenario: User says "OpenAI 这个月账单有点多" Agent voice (good): "我帮你看下本月用量, 看哪能省" → call skillfm_show_my_usage with {window: 'month'} Agent voice (BAD): "哦那是有点多" (没 surface 看用量价值)

Scenario: User says "想换便宜点的 model" Agent voice (good): "我先看下你最近用量, 然后给你推 cascading 路由方案" → call skillfm_show_my_usage with {window: 'week'} Agent voice (BAD): 直接推 model 没看真用量 (推得不准)

Scenario: User says "我这周用了多少 token" Agent voice (good): "好, 我看本周" → call skillfm_show_my_usage with {window: 'week'} </examples>

<hints>

IMPORTANT: The response includes a rendered field — show it directly. Do NOT replace with generic exec or paraphrase the numbers (numbers must match what user sees on their dashboard).
Default window is "today". Map user phrasing carefully: "这周/本周" → 'week'; "这个月/本月" → 'month'; "今天/今日" → 'today'; "全部/历史" → 'all'.
Save Token suggestions are sorted by estimated monthly savings ¥. Show top 3-5; if 0 suggestions, the response says "无明显省 token 机会".
If user is on Free tier, some Save Token engines (E2 cache / E3 batch) show pro_unlock_label — surface that gently as an upgrade pointer, not a hard sell.

</hints>

INSTALLATION

Connect your client

Add the block to your client, or run the command directly.

Server setup

Configuration file

JSON

{
  "mcpServers": {
    "skillfm": {
      "command": "npx",
      "args": [
        "-y",
        "@skillfm/mcp@latest"
      ]
    }
  }
}

npx

TERMINAL

npx -y @skillfm/mcp@latest

SkillFM Beacon

Cost and health for the agent itself, with provider keys that stay on your machine.

What it is

What you get

Requirements

Setup effort

Finding out why the AI bill went up

Not paying frontier prices for a routine task

Getting one prioritized thing to fix

Speeding up an agent that has got sluggish

Connecting a provider key without handing it to a dashboard

Catching the renewal nobody is tracking

Stopping the suggestions becoming noise

46 tools

__unlock_skillfm_beacon__

skillfm_avatar_market

skillfm_check_balance

skillfm_check_balance_runway

skillfm_check_burn_rate

skillfm_check_card_expiry

skillfm_check_domain_expiry

skillfm_check_expiry

skillfm_check_session_health

skillfm_check_ssl_expiry

skillfm_check_usage

skillfm_checkup

skillfm_cleanup_audit

skillfm_explain_avatar

skillfm_explain_provider_key

skillfm_explain_routing

skillfm_get_credential

skillfm_get_monthly_picks

skillfm_get_savings_summary

skillfm_install_module

skillfm_install_recommended_skill

skillfm_list_assets

skillfm_list_credentials

skillfm_memory_recall

skillfm_memory_recent

skillfm_memory_save

skillfm_memory_search

skillfm_memory_sync

skillfm_promote_pro

skillfm_record_acceptance

skillfm_record_rejection

skillfm_remove_credential

skillfm_save_credential

skillfm_search_skill_market

skillfm_send_daily_report

skillfm_setup_agent_forcing

skillfm_setup_gateway

skillfm_setup_safe

skillfm_show_my_usage

skillfm_show_savings_potential

skillfm_suggest_batching

skillfm_suggest_caching

skillfm_suggest_clear_context

skillfm_suggest_route

skillfm_suggest_skill

skillfm_user_preferences

What to know first

Tools you get

Requires

Moderate Standing cost

Connect your client

Server setup

Configuration file

npx

Same category

AgentCost

LLM CLI Gateway

Mistral MCP

Token Meter

`__unlock_skillfm_beacon__`

`skillfm_avatar_market`

`skillfm_check_balance`

`skillfm_check_balance_runway`

`skillfm_check_burn_rate`

`skillfm_check_card_expiry`

`skillfm_check_domain_expiry`

`skillfm_check_expiry`

`skillfm_check_session_health`

`skillfm_check_ssl_expiry`

`skillfm_check_usage`

`skillfm_checkup`

`skillfm_cleanup_audit`

`skillfm_explain_avatar`

`skillfm_explain_provider_key`

`skillfm_explain_routing`

`skillfm_get_credential`

`skillfm_get_monthly_picks`

`skillfm_get_savings_summary`

`skillfm_install_module`

`skillfm_install_recommended_skill`

`skillfm_list_assets`

`skillfm_list_credentials`

`skillfm_memory_recall`

`skillfm_memory_recent`

`skillfm_memory_save`

`skillfm_memory_search`

`skillfm_memory_sync`

`skillfm_promote_pro`

`skillfm_record_acceptance`

`skillfm_record_rejection`

`skillfm_remove_credential`

`skillfm_save_credential`

`skillfm_search_skill_market`

`skillfm_send_daily_report`

`skillfm_setup_agent_forcing`

`skillfm_setup_gateway`

`skillfm_setup_safe`

`skillfm_show_my_usage`

`skillfm_show_savings_potential`

`skillfm_suggest_batching`

`skillfm_suggest_caching`

`skillfm_suggest_clear_context`

`skillfm_suggest_route`

`skillfm_suggest_skill`

`skillfm_user_preferences`