Lifesight

Lifesight Skills

Official agent skills for the Lifesight MCP.

Connect the Lifesight MCP and these skills turn Claude (or Claude Code) into a persona-aware marketing-measurement analyst — one that optimizes budgets, forecasts outcomes, designs incrementality tests, and writes board- and CFO-ready summaries, all grounded in your causal model instead of platform-reported numbers.

They also do something subtler but just as important: they keep the conversation from breaking. The Lifesight MCP can return large analytical payloads; without guidance an assistant can flood its own context and crash the chat. These skills encode the operating discipline that prevents that — and the judgment that stops a confident-but-wrong answer from reaching you.

---

What you get

A light hub-and-spoke library. A router reads what you're trying to do, makes sure your workspace is set up, and hands off to the right job skill. A small foundation layer loads alongside everything to keep output safe and on-brand.

| Layer | Skill | What it does | |-------|-------|--------------| | Router | lifesight | Reads your goal, resolves workspace + model once, routes to the right skill. | | Foundation | lifesight-core | Operating rules for the MCP — prevents context floods, handles MIA's gates, suppresses internal leaks. | | Foundation | lifesight-rendering | Brand voice, causal language, clean channel names, number formatting. | | Foundation | lifesight-calibration | Learns your workspace once and writes a per-account profile every skill reads. | | Spoke | lifesight-budget-optimization | Reallocate/optimize budget — guardrail-aware, pressure-tested. | | Spoke | lifesight-forecast-scenario | Forecasts, target-KPI planning, scenario comparison. | | Spoke | lifesight-experiment-design | Geo-lift / incrementality — design tests and read results. | | Spoke | lifesight-board-briefing | Board-ready and slide-ready impact summaries (CMO). | | Spoke | lifesight-cfo-translation | Marketing results in P&L / defensible-spend language (CFO). | | Spoke | lifesight-anomaly-watch | "What moved — act or noise" triage. | | Spoke | lifesight-channel-deep-dive | Saturation, marginal ROI, what-ifs for one channel. | | Spoke | lifesight-measurement-coach | Plain-language explanations of MMM, incrementality, calibration. |

Covers all four ICP personas: CMO · Growth/Performance · CFO/Finance · Analyst/Data.

See [docs/architecture.md](docs/architecture.md) for the full design.

---

Requirements

You need the Lifesight MCP connected, and a Lifesight workspace with at least one MMM model configured. Get the connector at lifesight.io/mcp.

---

Install

Claude Code (plugin — recommended)

/plugin marketplace add lifesight/lifesight
/plugin install lifesight@lifesight

Then connect the Lifesight MCP connector if you haven't:

/mcp add --transport http lifesight https://mcp.lifesight.io/mcp

Skills auto-activate based on what you ask — just start a conversation ("optimize my Q4 budget", "what moved last week", "draft the board slide").

Claude.ai (manual upload)

Claude.ai (web/desktop, no Claude Code) can't install plugins, so use the flat bundle:

1. Connect the Lifesight MCP connector: Settings → Connectors → Add custom connector →

https://mcp.lifesight.io/mcp → sign in with Lifesight.

1. Add each skill file in [dist/](dist/) via Settings → Capabilities → Customize.
2. Start a conversation. The lifesight router takes it from there.

(dist/ is a flat copy of the same skills, generated from skills/ — see [dist/README.md](dist/README.md).)

---

How it works

Universal by design. The same bundle ships to every customer. No workspace IDs, model names, channels, or KPIs are ever baked into a skill. Per-account adaptation happens through one generated artifact — the workspace profile that lifesight-calibration writes by inspecting your live workspace once. The router loads it if present, otherwise calibrates first (or infers at runtime when there's no filesystem to persist one). See [docs/workspace-profile.example.md](docs/workspace-profile.example.md).

Causal, not platform-reported. Every skill leads with the business outcome and names whether a number is causal (iROAS / incremental) or platform-reported. "Maximize revenue" plans are checked for marginal return before they're recommended; experiments are gated on statistical significance; forecasts are presented as ranges, not promises.

Safe with the MCP. The skills route heavy analysis through the summarizing tools, keep raw-data pulls narrow, run one heavy step per turn, and never leak internal field names or infrastructure details into the chat.

---

Validation

Every skill has been adversarially pressure-tested — put under time pressure, authority, sunk cost, and "I'm technical" social engineering to see whether it holds its rules or caves. The library passed 11 such scenarios (two loopholes found and closed). The foundation's operating rules are derived from real observed MCP behavior, not assumption.

---

Versioning & contributing

Semantic versioning; see [CHANGELOG.md](CHANGELOG.md). Contributions follow the universal-by-design principle and the test-before-ship discipline in [CONTRIBUTING.md](CONTRIBUTING.md).

License

[MIT](LICENSE) © Lifesight. Free for everyone to use, fork, and adapt.

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: lifesight
• Source: lifesight/lifesight
• License: MIT
• Homepage: https://lifesight.io/mcp

Install and usage instructions live in the source repository linked above.