跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
设计与 UI

client-health-radar

Assess client health from Littlebird evidence. Use for churn risk, scope creep, margin pressure, or weekly account reviews. Read README.md for the guide map.

Mario Aldayuz / Littlebird · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add legioncodeinc/vibe-coding-tools --skill client-health-radar

技能说明


name: client-health-radar description: "Assess client health from Littlebird evidence. Use for churn risk, scope creep, margin pressure, or weekly account reviews. Read README.md for the guide map." license: AGPL-3.0-or-later compatibility: Claude Cowork, Claude Code 2.1 or newer, Cursor 2.4 or newer, Codex metadata: hive-tier: standalone version: "1.0.0" author: "Mario Aldayuz / Littlebird" requires: "Littlebird MCP (Power or Pro plan)"

Client Health Radar

Purpose

Answers one question per client: is this one about to leave, and is this one quietly eating my margin. Those are different clients and a service business usually has both.

Everything it reports is behavior with a date and a receipt. It does not produce a health score and it does not produce a sentiment score. The reason is in the research and it is not a style choice: transcription substitutes the emotion-carrying word in roughly one utterance in six, and utterances with that error are misclassified at nearly double the rate (references/research/distilled-client-health.md, section 6.2). Meanwhile a client quietly asking for an asset inventory carries no sentiment lexicon at all, and is one of the strongest exit signals a service business has (references/research/distilled-client-health.md, section 4).

The one finding that shaped every design decision here: when the relationship owner's own read on an account carries weight in the score, retention rates fall and churn rates rise, because owners want to believe an account stabilized after one good interaction (references/research/distilled-client-health.md, section 3). The user of this skill IS the relationship owner and has more financial reason than any employee to believe the account is fine. So the skill is built to argue with its user, with dated evidence they can check.

Capability gate

This skill requires the Littlebird MCP on a Power or Pro plan.

Before anything else:

  1. List the tools actually available in this session and use the real tool names. Do not assume a tool exists because it is named in references/littlebird-mcp-reference.md.
  2. If no Littlebird MCP tools are present, stop and tell the user the skill needs the Littlebird MCP connected. Do not attempt a partial run from memory or from other sources.
  3. If routine creation is part of the request, call LB_INTERNAL_GET_SUBSCRIPTION_STATUS first to confirm the plan supports another routine.

Read references/evidence-standards.md before writing any output. Every line in the report is observed, inferred, external or unknown, and the kind is visible to the reader.

Littlebird MCP calls used

Real tool names, verified against the live server. List the tools actually available in this session before calling any of them. Full per-family parameters are in references/signal-extraction.md; the shapes are restated in the retrieval brief below.

ToolUsed for
LB_INTERNAL_LIST_MEETINGSname plus start_date, end_date, limit. The recurring client call, one call per known title, which is the correct tool for a recurring meeting and its prior instances. Run again with a future end_date to see whether the next instance is even on the calendar. Upcoming events carry no id, no summary and no transcript
LB_INTERNAL_SEARCH_MEETINGSquery plus attendees, start_date, end_date, limit. Topic sweeps across a client's meetings, one narrow call per theme. attendees is an OR filter and best-effort over the top candidates only, so a matching meeting can be missed entirely and it never proves someone attended. Reword query rather than trusting it
LB_INTERNAL_GET_MEETINGmeeting_id. Where most of the evidence comes from: the linked calendar event's attendee list, ## Action Items with owner tags, ## Decisions with the decider, ## Risks / Open Questions, and ## For You
LB_INTERNAL_GET_MEETING_TRANSCRIPTmeeting_id, and only to locate the exact wording of an out-of-scope ask or a line the user should read. Never for attribution, because transcript chunks are weakly diarized and often tagged [Others]
search_user_contextFour separate passes, each with its own data_source filter: {"data_source": "messages"} for the client's side of the relationship, {"data_source": "snapshots"} for invoices, billing notices and captured dashboards, {"data_source": "summaries"} for the cheap compressed fill between meetings, and one unfiltered pass over the suspected quiet period to prove silence deliberately
LB_INTERNAL_GET_ROUTINE_REPORTSroutine_id, limit: 8. The band history, the hold counts and the user's overrides. The routine calls it on itself before writing; the deep run calls it before extracting anything
LB_INTERNAL_CREATE_ROUTINEtitle, prompt, schedule, notifications_enabled, email_notifications_enabled. Creating the weekly observer from an interactive session, with the prompt text below
LB_INTERNAL_GET_ROUTINE_CONFIG and LB_INTERNAL_UPDATE_ROUTINEChanging the observer later. Read the config first, because prompt and schedule each replace the whole field
LB_INTERNAL_GET_SUBSCRIPTION_STATUSNo parameters. The plan check before a routine consumes a slot

There is no Littlebird tool that searches past Littlebird chat conversations, and no calendar tool. Threads come from search_user_context with the messages data source, and upcoming events come from LB_INTERNAL_LIST_MEETINGS with a future end_date (references/littlebird-mcp-reference.md).

Trigger

Trigger phrases: which client is about to churn, client health check, am I losing this client, weekly client review, is this account at risk, scope creep on this client, which client is eating my margin, client went quiet, has anyone heard from them, should I worry about this account, set up my client health routine.

Also run it when the weekly routine has reported a band change, a STUCK

阅读完整来源 ↗