Brief To Shortlist

Brief-to-Shortlist

The flagship casting skill. Takes a raw client brief in any written format — forwarded email, deck text, scattered bullets, an RFP paragraph — and turns it into a ranked, enriched, client-ready shortlist that a casting director can paste into a deck without re-checking the data framing. For agency casting directors and brand-side influencer-marketing managers.

Read first: ${CLAUDE_PLUGIN_ROOT}/shared/conventions.md (tool schemas, credit prices, the conventions — conventions 10 budget-is-a-price, 11 brand filters, 12 relax the binding constraint, and 13 auto-chain profile lookups especially apply to casting). This skill honors thrifty/thorough credit modes (${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).

Calibration: if ~/.claude/plugins/config/creatorland/creatorland-data/PROFILE.md exists (written by /intake), read it first — persona, familiarity (1–3), and saved defaults set narration level and assumed context (conventions.md §15).

If the input is a call transcript or meeting recording rather than a written brief, this is the wrong skill — use transcript-to-shortlist. If the user wants a brief document produced (not creators found), use brief-builder.

Precision & filters for outreach (convention 14). When this list will feed an outreach step, default to precision over raw recall: search precision: "tight" with the hard-gate filters the brief supports (platform, niche, data_freshness_days, content_format, audience_country), and reserve broad plus heavy lookalike unioning for market sizing. Do not fan out dozens of lookalike calls to hit a volume target (each hop drifts from the seed), and trim the weak tail by each result row relative_fit (within-set fit, 1.0 = strongest) rather than padding to a round number. Canon: ${CLAUDE_PLUGIN_ROOT}/shared/conventions.md convention 14.

Lead with the supply answer when the client asks "is there enough?" Before (or alongside) casting, count_creators gives a free, rounded headline count of how many creators match the brief's filters — the client-facing "pool size" number. search_creators returns a capped ranked page, so quote the count for supply and the shortlist for the curated picks; never relay the shortlist length as the total pool.

Inputs to collect

• The brief (required) — any written format. Do not make the user restructure it; normalization is your job.
• Shortlist size — default 8 if unstated. Don't ask if the brief or context implies a size ("top 10", "a handful").
• Markets / geo priorities — pull from the brief if present; ask only if absent AND geo plausibly matters to the campaign.
• Competitor brands to conflict-check (optional) — ask once: "Any competitor brands I should conflict-check the slate against?" Skip if the user already named them or said no conflicts matter.
• Credit mode — default thorough; switch to thrifty on the trigger phrases in the credit-modes module.

Never ask twice for anything the brief already states — quote it back instead.

Flow

Step 0 — Normalize the brief (no tool call, 0 credits). Read the raw brief and produce a structured campaign description: target audience, corpus vertical (Beauty, Fashion, Health & Fitness, Food & Beverage, Technology, …), content style / vibe words, platform(s), geo markets, follower tier, explicit exclusions. Show it in one compact block; proceed unless the user corrects it. Keep the brief's own language — it powers the per-creator rationale later.

Two extraction rules to get right here (conventions 10–11):

• Budget is a price, not an audience size. If the brief gives money per deliverable or a campaign total ("$3–5K/video", "we have $20K"), record it as the rate band to check — never as a follower filter. Set follower filters ONLY from an explicit audience-size ask ("50–200K followers") or a named tier. Budget and tier are different axes; a budget alone never sets min_followers/max_followers.
• Named brands → a comp set. If the brief names brands to match on ("creators who've worked with similar beauty brands — I'm at Fenty"), build a comp set from the named brand and its peers (Fenty → Rare Beauty, Pat McGrath, Glossier, Charlotte Tilbury, …) and carry it into the search as a brand_affinities filter (step 1). Show the comp set in the normalized block so the user can correct it.

Alternative spine: the server ships a maintained prompt, shortlist-from-brief, that encodes this same flow server-side. Prefer it when the harness surfaces server prompts cleanly; otherwise run the steps below directly. Either way the deliverable template and honesty rules in this file still apply.

Step 1 — Search. One call, limit = 2–3× shortlist size (thorough: 3×; thrifty: 2×). For the default shortlist of 8, thorough:

search_creators {
  "mode": "brief",
  "brief": "<normalized brief text, ≤2000 chars, keeping the client's vibe words>",
  "filters": {
    "platform": "<if the brief names one: instagram|tiktok|youtube|twitter|twitch>",
    "country": "<primary market if stated>",
    "min_followers": <tier floor — ONLY from an explicit audience-size ask or named tier; NEVER inferred from a budget>,
    "max_followers": <tier ceiling — same rule; a "$3–5K/video" budget is a rate, not a size>,
    "niche": "<if cleanly stated>",
    "brand_affinities": ["<comp-set brands from step 0 when the brief names brands to match on>"],
    "min_engagement_rate": <0–1, only if the brief sets a bar>,
    "audience_country": "<ISO country ONLY when the ask is audience-in-market — e.g. 'reach US audiences' — distinct from where the creator lives>",
    "min_audience_country_share": <0–1, pair with audience_country when a minimum in-market share matters>,
    "data_freshness_days": <e.g. 90, ONLY when recency matters — 'creators active recently'>,
    "content_format": "<personality_led | faceless_clip — ONLY when the brief implies on-camera personality vs faceless/clip content>"
  },
  "limit": 24,
  "precision": "balanced"
}

Include only filters the brief actually supports — brief text alone is valid; if you pass filters it needs at least one signal. Use precision: "tight" when the brief is highly specific, "broad" when it's a vibe sketch.

When the brief names brands to match on, pass the step-0 comp set as brand_affinities (and confirm via profile brand affiliations in step 4) — more reliable than hoping brief text surfaces brand-adjacent creators semantically (convention 11). Never translate the campaign budget into min_followers/max_followers (convention 10) — the budget is handled by the rate call in step 5, not the search filters.

The GA hard-gated filters (conventions §search_creators) are advisory, not mandatory — reach for them only when the brief calls for it: audience_country (+ min_audience_country_share) when the ask is audience-in-market (distinct from where the creator is based), data_freshness_days when recency matters, and content_format when the brief implies personality-led vs faceless/clip content. Each is a hard gate (filters the pool, no rank score), so don't over-constrain — if a gate empties the pool, relax it per convention 12 and say so.

If results come back thin, relax per convention 12 — broaden the binding constraint, not a satisfied one: if the user said "California", widen geo at the same level (California → West Coast / US) and offer audience-geo as an alternative axis; never suggest sub-regions already inside the stated region, and never propose switching off the requested platform. Say which constraint you relaxed and why.

Step 2 — Profile fan-out. get_creator_profile per candidate:

get_creator_profile { "identifier": { "type": "creatorland_user_id", "creatorland_user_id": "<id from search result>" } }

(Use whichever identifier type the search result returns — social_handle, creatorland_user_id, or source_user — exactly one type per call.) Thorough: profile every candidate from step 1. Thrifty: rank candidates on search-result signal first, profile only the top shortlist size + 2.

On paid plans (pro/pilot/internal) both the search rows and each profile carry a paid-gated avatar { url, source }; free/demo plans get avatar: null. Keep the URL against each creator so the shortlist card can show a headshot — degrade gracefully to initials when it's null.

Credit estimate fires here. Thorough at default size is 2 + 24 + 5 ≈ 31 credits — over the ~30 threshold, so state the estimate before fanning out and offer thrifty (do not block; proceed after stating it).

Step 3 — Freshness Gate. Classify every profiled creator fresh / aging / stale per the freshness-gate module. Stale creators go only in the "re-verify before pitch" section (thrifty: may drop them, saying how many were dropped).

Step 4 — Conflict check. If the user supplied competitor brands, compare each profile's brand affiliations (pro-plan field) against that list. Overlap → a conflict flag on the row, naming the brand and that it came from the profile's affiliations. If affiliations are absent (free plan), say conflict checking was unavailable and why — never silently skip it.

Step 5 — One rate call for the slate. Exactly one query_market_intelligence for the whole slate's vertical — shared band context, not per-creator pricing:

query_market_intelligence {
  "mode": "rate",
  "vertical": "<the slate's inferred vertical>",
  "deal_type": "<only if the brief specifies one>",
  "creator_tier": "<the slate's target tier — if the brief names one or the finalists cluster at one>"
}

Wrapped in Refusal Recovery: if refused (rate floor: 10 brands / 50 deals), walk the ladder (thorough: until clearance; thrifty: max 2 rungs) and disclose the clearance level in the deliverable. Each rung is a fresh 5-credit call.

When the brief names a creator tier (or the finalists cluster at one), pass creator_tier (emerging <1k / nano 1k-10k / micro 10k-100k / mid 100k-500k / macro 500k-1M / mega 1M+) so the band is scoped to same-size creators; when tier is mixed or unknown keep the vertical-wide band as the fallback and say so — a tier too thin for the distinct-creator privacy floor auto-broadens (disclosed) to the all-tier band.

If the brief stated a budget, this rate band is how that budget is honored (convention 10): surface the band up front and flag which finalists sit inside the stated budget — appropriately priced ✓, above band ↑, below band ↓ — as a rate-fit note on each row. Never disqualify a creator for being "too small" for the budget; the budget is spend, not an audience filter.

Step 6 — Rank and write the deliverable. Rank on brief-fit (audience, vibe, geo, tier), penalizing aging freshness and conflicts. Tie rationale to the brief's own language.

Deliverable

Client-ready markdown (paste-into-deck quality):

# Creator Shortlist — <campaign/brand name>
_Built from your brief, <date> · <N> creators · Creatorland Data_

## The brief, as I read it
<the normalized block from step 0, 3–5 lines>

## Shortlist (ranked)

### 1. <Name> — @<handle> (<platform>, <follower count>)
<!-- when avatar.url is present (paid plans), render the headshot on the card; fall back to initials when null/free -->
- **Why for this brief:** <2–3 sentences quoting the brief's language — "you asked for 'lo-fi kitchen energy'; her last-90-day grid is exactly that">
- **Audience-geo fit:** <top geos vs. the brief's markets, e.g. "62% US / 14% UK — matches your US-first ask">
- **Rate fit:** ✓ in your $<budget> band | ↑ likely above band | ↓ likely below band — vs. the slate rate band below (only when the brief gave a budget)
- **Brand fit:** <when the brief named brands to match on: "affiliations include <comp-set brand>" or "no comp-set affiliation found in corpus">
- **Freshness:** fresh | aging (note) — stale creators never appear here
- **Conflicts:** none found | ⚠ affiliated with <brand> (per profile brand affiliations) — review before pitch
<repeat per creator>

## Indicative rate context (slate-level)
Market band for the **<vertical>** vertical<, broadened per note below>:
p25 $<x> · median $<y> · p75 $<z> — <deal volume / recency from the tool>.
> <provenance line exactly as the tool returned it>
This is the **market band for the vertical/tier — not any individual
creator's rate.** Use it to frame budget conversations, not to quote talent.
<If broadened: "Benchmark basis: <level> — the narrower slice was below the
privacy floor (10 brands / 50 deals), so scope was widened; a privacy feature
of the data source, not missing data.">

## Re-verify before pitch
<stale creators with what's stale; or "None — all candidates cleared the freshness gate."
Thrifty + drops: "K stale candidates were dropped in thrifty mode.">

## Caveats
- Rate band is vertical-level; no per-creator rate data exists in this corpus.
- Conflict checks cover Creatorland-known brand affiliations only — not
  unreported or in-flight deals.
- <conflict checking unavailable on free plan, if applicable>
- <any filters the corpus couldn't honor — report gaps, don't paper over them>
- No creator contact info is included or available via this tool; route
  outreach through Creatorland connections or the creator's public profiles.

---
Data freshness: <N>/<M> creators synced within the last sync window; <K> flagged for re-verification.
Provenance: Creatorland Data MCP · search + <M> profiles + 1 rate benchmark (<clearance level>) · <date>.
Credits used this run: ~<N> (1 search ×2 + <M> profiles ×1 + <R> market-intel ×5).

Honesty rules

• The rate band is vertical-level, never "this creator's rate." Write "market band for this vertical/tier" — no per-creator pricing exists.
• Disclose Refusal Recovery broadening at the clearance level, always.
• Never strip the provenance line to make the deck prettier.
• Rationale must trace to brief language or profile data — no invented audience claims, no engagement numbers the tools didn't return.
• Stale data is never silently mixed into the main list.
• A stated budget is spend, not an audience size — never disqualify a creator for being "too small/large" for it, and never set follower filters from it (convention 10). Rate fit is framed against the slate's vertical band, never as that creator's actual rate.
• When you relax a thin search, never suggest a sub-region already inside the user's stated region, and never propose abandoning the requested platform (convention 12); name the constraint you relaxed.
• Never imply access to creator contact info (PII discipline, convention 7).

Credit footprint

thorough: ~31 credits (1 search ×2 + ~24 profiles + 1 rate call ×5; +5 per refusal-ladder rung) · thrifty: ~17 credits (1 search ×2 + ~10 profiles + 1 rate call ×5, max 2 ladder rungs)