Campaign Monitor

Campaign Monitor

A live campaign needs a living dashboard. This skill takes the campaign's rostered creators, pulls each one's OAuth-consented, account-level stats via get_member_content_stats, frames the numbers against the vertical's market band, and assembles a per-creator dashboard — with honest, labeled slots for creators who aren't members or haven't connected socials, so coverage is visible rather than hidden. It is the in-flight monitoring artifact; for the post-campaign wrap deck use wrap-report-skeleton. For agency account leads and brand-side managers running a campaign right now.

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), Refusal Recovery (${CLAUDE_PLUGIN_ROOT}/shared/refusal-recovery.md), and the Freshness Gate (${CLAUDE_PLUGIN_ROOT}/shared/freshness-gate.md).

The new tool this skill wraps

get_member_content_stats — 1 credit per call (per-row, like a profile fan-out). Newly live as of CRE-541; not yet in shared/conventions.md, so its shape is documented here:

• Input: { identifier: { type, ... } } — exactly one of the same five identifier types as get_creator_profile: social_handle, creatorland_user_id, source_user, email, phone.
• Success envelope (connected): consented: true, providers[], plus accounts[] where each account carries provider, username, is_verified, followers, subscriber_count, view_count, likes_count, video_count, media_count, engagement_signal (likes/views, or null), source_updated_at; plus a top-level data_as_of and freshness.
• Refused / not-connected envelope (this is a SUCCESS, not an error): the tool returns a clean refused/not-connected envelope — never an exception — for (a) a creator who isn't a Creatorland member, (b) a member with no connected/OAuth'd socials, or (c) a plan without the entitlement. Treat each of these as a labeled dashboard slot, not a failure.
• Scope ceiling (hard): this tool is account-level only. It returns per-provider account totals. Per-post / per-video engagement and audience demographics do NOT exist yet (deferred to CRE-653). This skill must never fabricate per-post metrics, per-deliverable performance, or audience breakdowns — if asked, say they're not available yet and point to the account-level figures that are.

Inputs to collect

• Campaign roster (required) — the creators on the live campaign, in any format (handles, emails, a CSV, or a prior shortlist/slate from this session). Map each row to one of the five identifier types (a handle column → social_handle, an email column → email, a Creatorland ID → creatorland_user_id, etc.). If rows mix types, map each row to its own type. Ask which column is the identifier only if ambiguous.
• Campaign vertical (required for the benchmark band) — infer from an in-session brief/shortlist if one exists; otherwise ask once. Sets the query_market_intelligence filter.
• Deal type (optional) — sharpens the benchmark band if the corpus supports it.
• Refresh cadence (optional) — if the user wants this to be a standing dashboard, note the cadence so the schedule tip fires (see Flow step 5).
• Mode — default thorough; thrifty on the standard triggers.

Never re-ask for identifiers or a vertical the conversation already holds.

Flow

Step 0 — Count rows and estimate (mandatory over ~30 credits). The stats fan-out is 1 credit per roster row, plus ~5 per benchmark call, so the pre-run credit estimate is mandatory once the roster pushes the plan past ~30 credits (and stated regardless once N×1 + benchmark > 30):

Estimated cost: <N+5> credits ($<(N+5)×0.025>) — content-stats calls (1 each) + 1 market-intel benchmark (5). Say "thrifty" to monitor only the connected creators you name, or cap the row count.

State it, then proceed (do not block). Thrifty: pull stats only for a user-named priority subset (and still surface the un-pulled rows in the coverage summary so they aren't hidden), and take at most 2 refusal-ladder rungs on the benchmark.

Step 1 — Per-creator stats fan-out — for each roster row: get_member_content_stats { "identifier": { "type": "<row's type>", ... } } (exactly one identifier type per call, matching that row). Classify the envelope into one of three consent slots — this is the heart of the skill:

• 🟢 Connected — consented: true with accounts[]. Capture per provider: followers / subscriber_count, view_count, likes_count, video_count / media_count, engagement_signal, is_verified, and source_updated_at. These rows carry real numbers.
• 🟡 Member, no connected socials — refused/not-connected envelope for a member who hasn't OAuth'd any socials. Labeled slot: "member, no socials connected — no consented stats to show." Never an error, never blank- implying-zero.
• ⚪ Not a Creatorland member — refused envelope for a non-member. Labeled slot: "not a Creatorland member — outside the consented-stats surface."

A plan-without-entitlement envelope is labeled "stats unavailable on plan", not blank. 1 credit per call regardless of which slot it lands in.

Step 2 — Freshness Gate — classify each connected row from its freshness / source_updated_at: fresh / aging / stale. Stale rows keep their numbers but carry the "figures may have drifted — re-verify" caveat inline (you can't exclude a live-campaign creator; you flag the drift). The 🟡/⚪ slots are not freshness-gated — they have no figures to age.

Step 3 — Benchmark framing — one query_market_intelligence per distinct campaign vertical (usually one), wrapped in Refusal Recovery (market-mode floor 5 brands / 25 deals; rate-mode floor 10 brands / 50 deals): { "mode": "market", "vertical": "<campaign vertical>" } for the activity/ engagement context band, and in thorough mode also { "mode": "rate", "vertical": "<vertical>", "deal_type": "<if known>", "creator_tier": "<the monitored roster's tier, if it shares one>" }. When the connected roster shares a creator tier, pass creator_tier (emerging <1k / nano 1k-10k / micro 10k-100k / mid 100k-500k / macro 500k-1M / mega 1M+) so the band is size-scoped; when tiers are mixed keep the vertical-wide band as the fallback (a too-thin tier auto-broadens, disclosed). This gives each creator's account-level numbers a "vs market" band so a follower or engagement figure reads against the vertical norm, not in a vacuum. Walk the ladder on refusal, stop at first clearance, and disclose the clearance level in the footer ("a privacy feature of the data source, not missing data"). 5 credits per call incl. retries. Thrifty: max 2 rungs; market-band call only.

Step 4 — Assemble the dashboard (no tool calls) — build the artifact below: per-creator rows for connected creators, the labeled 🟡/⚪ slots, roster totals across connected accounts, a consent-coverage summary, and the benchmark band. Emit it as a harness-side dashboard artifact/canvas where the harness supports one (the way the deck/dashboard skills do — a single markdown document the user can keep open and re-render); otherwise inline markdown. Re-running the skill overwrites the same artifact so it reads as a living dashboard, not a new file each time.

Step 5 — Scheduled refresh (make it living) —

Tip: in a harness with scheduled tasks (e.g. Cowork scheduled tasks), schedule "refresh the campaign-monitor dashboard for ''" daily or weekly. Each run re-fans-out the stats, re-pulls the benchmark, and overwrites the same dashboard artifact, so it stays current for the length of the flight. On-demand runs are the baseline; never depend on scheduling existing.

Deliverable

A campaign performance dashboard in markdown (single artifact, re-renderable):

# Campaign Monitor — <campaign / brand name>
_Live dashboard · <N> rostered creators · data as of <data_as_of> · <date>_

## Consent coverage
🟢 <c> connected · 🟡 <m> member, no socials · ⚪ <u> not a member
<c>/<N> creators reporting consented stats. Un-reporting slots are shown
below, not hidden — coverage is a campaign fact, not an error.

## Connected creators (account-level, consented)
| Creator | Provider | Followers/Subs | Views | Likes | Videos/Media | Engagement signal | Verified | Freshness |
|---|---|---|---|---|---|---|---|---|
| @<handle> | instagram | <n> | <view_count> | <likes_count> | <count> | <likes/views or —> | ✓/— | fresh/aging/⚠ stale |
<one row per connected account; a multi-platform creator gets one row per provider>

## Not reporting (labeled, never blank)
| Creator | Status |
|---|---|
| @<handle> | 🟡 member, no socials connected — no consented stats |
| @<handle> | ⚪ not a Creatorland member — outside the consented-stats surface |
<or "All rostered creators are connected.">

## Roster totals (connected accounts only)
Followers/subs across connected accounts: <Σ> · views: <Σ> · likes: <Σ>.
<Totals cover the <c> connected creators only — the 🟡/⚪ slots contribute
no figures by design.>

## Benchmark band (vs market)
Market context for the **<vertical>** vertical<scope note if broadened>:
<activity / engagement band from market mode; rate band p25 $<x> · median
$<y> · p75 $<z> if pulled>. Read each creator's account-level numbers
against this corpus band — it is the vertical norm, not any creator's
personal history.
> <provenance line verbatim from the tool> · recency window: <window>

## Not available yet
Per-post / per-video engagement and audience demographics are **not exposed
yet** (CRE-653). This dashboard is account-level only; these slots are left
out deliberately rather than estimated.

---
Consent coverage: <c>/<N> connected; <m> member-no-socials; <u> non-member.
Data freshness: <F>/<c> connected creators synced within the last sync
window; <K> flagged — account figures may have drifted.
Benchmark basis: <clearance level; if broadened: "a privacy feature of the
data source, not missing data">.
Provenance: Creatorland Data MCP · <c> content-stats calls + <R> market-intel · <date>.
Credits used this run: ~<N+5R> (<N> content-stats ×1 + <R> market-intel ×5).

Honesty rules

• Account-level only — never fake per-post. get_member_content_stats returns per-provider account totals; per-post / per-video engagement and audience demographics do not exist yet (CRE-653). Never estimate, infer, or imply them — the "Not available yet" block says so explicitly.
• Un-consented slots are visible, not hidden. The refused/not-connected envelope is a SUCCESS; every 🟡/⚪ creator appears as a labeled slot. An honest dashboard shows its own coverage holes — that is the feature.
• Consent is the data source. Stats are first-party, OAuth-consented, per-provider — never scraped, never inferred from a public profile. Say so.
• Freshness is disclosed per connected row from source_updated_at / freshness; stale rows keep numbers but carry the drift caveat (live creators can't be excluded).
• Benchmark is corpus-level for the vertical — "market band for the vertical", never "what these creators normally do" (convention 2). Disclose every Refusal Recovery broaden at the clearance level.
• Plan honesty. A no-entitlement envelope is labeled "stats unavailable on plan", never blank-implying-zero.
• PII invariant (convention 7). No contact info in the dashboard, regardless of source — including any contact data in the uploaded roster. The tool returns none; you add none.

Cross-references

• Casting the campaign (building the roster in the first place) → brief-to-shortlist.
• Post-campaign wrap deck (flight is over, reporting is due) → wrap-report-skeleton.
• Enriching the roster with profile data (interests, geo, affiliations) → roster-enricher.
• Benchmarking the roster's rates rather than performance → rate-card-generator.

Credit footprint

thorough: ~(N + 5–10) credits for an N-creator roster (N content-stats ×1 + 1–2 market-intel ×5; e.g. 15 creators ≈ 20–25) · thrifty: ~(subset + 5) credits (named-subset content-stats only + 1 market-band call, max 2 ladder rungs), with un-pulled rows still shown in the coverage summary. Estimate stated up front whenever the plan exceeds ~30 credits.