Triangulated Casting

Triangulated Casting

The highest-confidence casting move on the surface: combine both search_creators modes. Run brief mode on the campaign requirements AND lookalike mode seeded from a creator who already performed well, then intersect. Creators that appear in BOTH searches are the strongest picks — they match the stated brief and resemble proven talent. For casting directors who have a past winner and want "more of that, but on-brief."

Read first: ${CLAUDE_PLUGIN_ROOT}/shared/conventions.md (tool schemas, credit prices, the nine conventions). This skill honors thrifty/thorough credit modes (${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).

This skill requires BOTH a brief and a seed creator. Brief only → brief-to-shortlist. Seed only, tiered by price → lookalike-ladder. The intersection is the whole point here.

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.

Inputs to collect

• The brief (required) — any written format; normalization is your job.
• The seed performer (required) — one creator who already worked well (handle/URL), or one standout post URL. Exactly one seed (lookalike takes one of seed_creator or seed_content). Multiple proven performers → run the skill once per seed and present as a slate (convention 9), or pick the single best with the user.
• Shortlist size — default 8.
• Competitor brands to conflict-check (optional) — ask once if unstated and conflicts plausibly matter.
• Credit mode — default thorough; thrifty on trigger phrases.

Flow

Step 0 — Normalize the brief (no tool call). Structured campaign description: audience, vertical, vibe words, platform, geo, tier, exclusions. Show it back; proceed unless corrected. Keep the client's language.

Step 1 — Brief search. One call:

search_creators {
  "mode": "brief",
  "brief": "<normalized brief text, ≤2000 chars, vibe words kept>",
  "filters": {
    "platform": "<if named>",
    "country": "<primary market if stated>",
    "niche": "<if clean>",
    "min_followers": <tier floor if set>,
    "max_followers": <tier ceiling if set>
  },
  "limit": <3× shortlist size (thorough) | 2× (thrifty)>,
  "precision": "balanced"
}

Step 2 — Lookalike search. One inference-free call — exactly ONE of seed_creator or seed_content, never both, no filters argument:

search_creators {
  "mode": "lookalike",
  "seed_creator": { "type": "social_handle", "platform": "<platform>", "handle": "<seed handle>" },
  "limit": <3× shortlist size (thorough) | 2× (thrifty)>,
  "precision": "balanced"
}

or "seed_content": { "url": "<standout post URL>" }.

Step 3 — Intersect (no tool call). Match the two result sets on creator identity. Bucket: in-both (the high-confidence core), brief-only, lookalike-only. The in-both set is the headline; the singles are depth. If the intersection is empty or tiny, say so plainly — it's a real signal that the brief and the proven performer point in different directions, and the user should choose which to trust. Never manufacture overlap.

Step 4 — Profile fan-out. get_creator_profile by returned identifier:

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

Thorough: profile the full in-both set plus enough of each single bucket to fill the shortlist. Thrifty: profile the in-both set plus the top singles only, up to shortlist size + 2.

Credit estimate fires here. Thorough is two searches plus a sizable fan-out — typically ~35–45 credits — over the threshold, so state the estimate and offer thrifty before fanning out (don't block).

Step 5 — Freshness Gate. Classify fresh / aging / stale; stale go to "re-verify before pitch" (thrifty: may drop, noting counts).

Step 6 — Conflict check (if competitor brands supplied). Compare each profile's brand affiliations (pro field) to the list; flag overlaps naming the brand and that it came from affiliations. Free plan / affiliations absent → say conflict checking was unavailable and why; never silently skip.

Step 7 — One rate call for the vertical. Exactly one query_market_intelligence, wrapped in Refusal Recovery:

query_market_intelligence { "mode": "rate", "vertical": "<the slate's vertical>", "creator_tier": "<the slate's target tier, if it clusters at one>" }

Walk the ladder on refusal (thorough: until clearance; thrifty: max 2 rungs); disclose clearance level. When the slate clusters at a creator tier, pass creator_tier (emerging <1k / nano 1k-10k / micro 10k-100k / mid 100k-500k / macro 500k-1M / mega 1M+) for a size-scoped band; when tier is mixed or unknown keep the vertical-wide band as the fallback and say so (a tier too thin for the privacy floor auto-broadens, disclosed, to the all-tier band).

Step 8 — Rank. In-both creators rank above singles by construction; within each bucket, rank on brief-fit and freshness, penalizing conflicts.

Deliverable

# Triangulated Shortlist — <campaign/brand name>
_Brief × lookalikes of @<seed>, <date> · <N> creators · Creatorland Data_

## The brief, as I read it
<normalized block, 3–5 lines> · **Proven seed:** @<seed> (<why they worked,
if the user said>).

## High-confidence core — in BOTH searches
### 1. <Name> — @<handle> (<platform>, <follower count>) — ★ in both
- **Why high-confidence:** matched the brief AND resembles @<seed>
  (embedding-based, inference-free) — <2–3 sentences tying to brief language>
- **Audience-geo:** <top geos> · **Freshness:** fresh | aging (note)
- **Conflicts:** none | ⚠ affiliated with <brand> (per profile affiliations)
<repeat for the in-both set; if empty: "No creator surfaced in both searches —
the brief and @<seed> point in different directions. Pick which signal leads;
the single-source picks below show each direction.">

## Brief-matched (brief search only)
<rows, same format, ranked>

## Lookalikes of @<seed> (lookalike only)
<rows, same format; vibe match is embedding-based, inference-free>

## Indicative rate context (slate-level)
Market band for the **<vertical>** vertical<, broadened per note>:
p25 $<x> · median $<y> · p75 $<z> — <deal volume / recency>.
> <provenance line exactly as the tool returned it>
Market band for the vertical/tier — **not any individual creator's rate.**
<If broadened: disclose clearance level.>

## Re-verify before pitch
<stale creators; or "None.">

## Caveats
- "High-confidence" = appears in both searches; it is not a performance
  guarantee. Lookalike similarity is embedding-based (the data's read of
  @<seed>), inference-free.
- Conflict checks cover Creatorland-known affiliations only.
- Rate band is vertical-level; no per-creator rate data exists.
- No creator contact info is included or available via this tool; route
  outreach through Creatorland connections or the creators' public profiles.

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

Honesty rules

• An empty/tiny intersection is reported honestly as divergence between brief and seed — never papered over by loosening the match.
• "High-confidence" means in-both, not guaranteed performance; lookalike similarity is embedding-based and inference-free.
• Disclose Refusal Recovery broadening at the clearance level, always.
• The rate band is vertical-level, never per-creator.
• Stale data never silently mixes into a bucket.
• Never imply access to creator contact info (convention 7).

Credit footprint

thorough: ~42 credits (2 searches ×2 + ~33 profiles + 1 rate call ×5; +5 per refusal rung) · thrifty: ~21 credits (2 searches ×2 + ~12 profiles + 1 rate call ×5, max 2 rungs)