Usage Pulse

Usage Pulse / CS Health Check (internal)

The anti-churn nudge. For a single customer account, this internal CS skill reads where the account is — its plan tier, what it's actually been running, where it's leaving value on the table — and produces a short, warm "3 things to try this week" email with copy-paste prompts, so a CSM can send activation in five minutes instead of writing it cold. The job-to-be-done: turn a quiet account into an active one before it lapses. The artifact is a per-account health read plus a ready-to-send email.

v1 capability note. There is no account-usage telemetry tool on the MCP yet (it's wishlist item 5, get_account_usage, tracked in CRE-543). Until it lands, this skill takes the account's current usage as an operator-supplied input — pasted from the billing/admin dashboard, an export, or the CSM's notes — and does not fabricate usage it cannot see. When the telemetry tool ships, step 1 below becomes a tool call and the rest is unchanged.

Read first: ${CLAUDE_PLUGIN_ROOT}/shared/conventions.md (tool schemas, credit prices, the conventions). This skill honors thrifty/thorough credit modes (${CLAUDE_PLUGIN_ROOT}/shared/credit-modes.md) and Refusal Recovery (${CLAUDE_PLUGIN_ROOT}/shared/refusal-recovery.md). It is near-zero-credit: it reasons over the skill catalog and the account's profile, and makes at most one optional market-intel call to season a prompt with a live number.

Inputs to collect

• Account (required) — name/segment (agency, brand, talent manager) and plan tier (free / pro / pilot). Tier gates which skills are even available, so it shapes every recommendation.
• Current usage (required for v1, operator-supplied) — what the account has run recently: which skills/workflows, rough call volume, credits consumed vs granted. Paste from the billing/admin dashboard. If the user can't supply it, say plainly that v1 can't read usage itself and proceed on tier + segment alone, clearly labeling the read as "tier-based, not usage-based".
• Their vertical / who they cast for (optional) — lets prompts be pre-filled with the account's real vertical instead of a placeholder.
• Known goals or friction (optional) — anything the CSM knows (e.g. "they only ever run shortlists, never pricing"). Sharpens the gap analysis.
• Mode — default thorough; thrifty skips the optional live-number call.

Flow

This skill is mostly reasoning, not tool calls.

1. Read the account (v1: from the operator-supplied usage; future: a get_account_usage call). Establish tier, what's been used, what's been granted vs consumed. No credits in v1.

2. Map the opportunity (no tool calls) — diff the account's actual usage against the catalog of skills its tier unlocks but it isn't using. Weight by fit to the account's segment and vertical:

   • Agencies idle on brief-to-shortlist/cross-market-slate -> casting depth.
   • Brands not touching vertical-briefing/competitor-watch -> market intel.
   • Anyone on pro never running pricing -> fair-price-brief/rate-card-generator. Pick the 3 highest-leverage unused workflows for this account — concrete, not the whole menu.

3. Season one prompt with a live number (thorough only, optional) — at most one query_market_intelligence { mode: "rate" | "market", vertical: <their vertical> } (5 credits, wrapped in Refusal Recovery) — in rate mode you may optionally add creator_tier (emerging <1k / nano 1k-10k / micro 10k-100k / mid 100k-500k / macro 500k-1M / mega 1M+) for a tier-specific hook, or omit it (the default) for the vertical-wide band — so a recommendation can carry a real "median rate in your vertical is $X — here's the skill that uses it" hook. Skip in thrifty. Never imply this number is account-specific; it's a vertical market band (convention 2).

4. Write the email (no tool calls) — warm, short, 3 concrete suggestions, each with a one-line "why it helps you" and a paste-ready prompt the account can drop straight into their session. Respect tier: never recommend a workflow their tier doesn't unlock — instead, if the highest-value next step is gated, frame it as the upgrade reason, honestly.

Credit footprint is ~0–5 credits total (one optional market call). No pre-run estimate needed.

Deliverable

# Usage pulse — <Account> (<tier>, <segment>)
*Prepared <date> · read basis: <"usage dashboard, <date>" / "tier-based only —
usage not supplied">*

## Health read
- Using well: <skills/workflows they already run>
- Leaving on the table: <tier-unlocked workflows they aren't using>
- Risk note: <e.g. "low credit consumption vs grant — activation risk"; only if
  usage was supplied>

## The 3 to try this week
1. **<Skill / workflow>** — <one line why it fits this account>.
2. **<Skill / workflow>** — ...
3. **<Skill / workflow>** — ...

## Ready-to-send email
> Subject: 3 quick wins in Creatorland this week
>
> Hi <name>,
> <2-3 warm sentences.>
>
> 1. **<Win>** — <why>. Try this:
>    `<paste-ready prompt, pre-filled with their vertical>`
> 2. ...
> 3. ...
>
> <Soft close + offer to walk through one live.>

---
**Read basis:** <usage source + date, or the honest "tier-based, usage not
read in v1 — no account-telemetry tool yet (CRE-543 wishlist item 5)">.
**Provenance (if a live number was used):** <the market-intel provenance line,
verbatim, + recency window — and it is a vertical market band, not this
account's data>.
Credits used this run: ~N (0 if thrifty / no live number; 5 if one market call)

Honesty rules

• Don't invent usage. v1 has no telemetry tool. If usage wasn't supplied, the read is explicitly "tier-based, not usage-based" — never a fabricated "you've run X searches". Say what you can and can't see.
• Respect the tier. Never recommend a workflow the account's tier doesn't unlock as if it were available; if it's the best next step, frame it as an honest upgrade reason, not a bait.
• Live numbers are vertical market bands (convention 2) — a "$X median in your vertical" hook is corpus-level context, never the account's own rate or a creator's rate.
• Provenance survives into the email when a live number is used (convention
  1. — the paste-ready prompt or a footer carries the citation.
• No contact info in the deliverable (convention 7); the email is to the account, but creator/brand contact data never appears in it.
• Internal artifact, honest framing. This is a CS tool; the activation suggestions are genuine fit reads, not upsell padding — recommend what actually helps the account, in priority order.

Credit footprint

thorough: ~0-5 credits (catalog reasoning is free; one optional market-intel call is 5, +5 per ladder retry) · thrifty: 0 credits (no live number).