Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when the user asks to "create an influencer brief" or "write a campaign brief"; produces a structured creator brief with deliverables, key messages, creative direction, timeline, disclosure rules, and compensation terms. Not for choosing how to split spend across creators — use budget-optimizer. 达人合作简报/创作者BF
.claude/skills/aaron-he-zhu-brief-generator/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 42% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 68% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 11% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 95% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 152% | 0% |
This skill helps you create clear, comprehensive influencer briefs that set creators up for success. Good briefs lead to better content, fewer revisions, and stronger partnerships.
Shortest invocation:
Create an influencer brief for [campaign]Common scenario:
Generate a TikTok brief for micro-influencers promoting [product], 1 review video, with disclosure and timelinecreator_ref, brand_ref, page_ref, shipping_ref, contact_ref, brand_asset_ref, hashtag_ref, promo_code_ref, and (when voice is used) voice_source_ref; plus memory/projections/narrative.json, memory/projections/claims.json, and relevant creator/channel projections. HOT is only an index to those sources. Raw creator/brand names and handles, page/asset-folder URLs, shipping addresses, contact names/emails/phones, hashtags, promo codes, and voice-source locators are transient rendering/dispatch inputs only.memory/influencer/brief-generator/YYYY-MM-DD-<topic>.md only with exact authorization. Saved artifacts and handoffs represent creator, brand, landing destination, shipping destination, contact path, assets, hashtags, promo terms, and voice provenance only with creator_ref, brand_ref, page_ref, shipping_ref, contact_ref, brand_asset_ref, hashtag_ref, promo_code_ref, and voice_source_ref; never persist their raw identity/address/URL/contact/content values or a hidden mapping. Each unresolved-claim proposal is a separate exact operation: propose authorization; a brief-save approval does not cover it.> Emit the standard shape from skill-contract.md §Handoff Summary Format, including the Narrative/claims dependency tuple.
Required fields: narrative_canon_id, narrative_canon_version, claims_projection_offset, and dependency_status: verified | approved-fallback | blocked, plus the applicable opaque creator_ref, brand_ref, page_ref, shipping_ref, contact_ref, brand_asset_ref, hashtag_ref, promo_code_ref, and voice_source_ref values. Do not include their raw resolved values.
This family has no live integrations required (Tier 1). The skill works end to end by asking the user for inputs: campaign details, deliverables, key messages, timeline, and compensation. Provide those in the prompt and you get a complete brief with zero setup.
Optional connectors that can enrich a brief when available:
~~influencer database — resolve creator details transiently for personalization while retaining only creator_ref and opaque evidence refs in the saved brief/handoff.~~social platform analytics — confirm current format specs and best-performing post lengths per platform.~~CRM — fetch the assigned contact_ref and prior brief versions; resolve raw contact details only during an authorized dispatch.Read accepted Narrative and claims projections before drafting. Claim approval is contextual: audience, market, media, offer window, and required disclaimer must match. No usable canon permits only an explicitly approved exploratory brief, never a creator-ready/on-canon label.
See CONNECTORS.md for the verified free/keyless recipe per category. None are required.
When a user requests a brief:
creator_ref, brand_ref, page_ref, shipping_ref, contact_ref, and voice_source_ref; resolve HOT pointers to their actual source records. Read Narrative/claims projections at named offsets. Keep raw identity, page, address, contact, and voice-source locators transient. If creator voice is required, capture the reference-safe intake via creator-voice-intake.md.[needs source], offer an exact registry-events.py operation: propose request for separate authorization, and prevent creator-ready status until resolved.Send, “deliver,” or “share with creators” request/label creates only a pending outreach handoff; it does not authorize delivery. Hand the brief to outreach-manager, resolve raw recipient/brand/page/contact values only transiently inside the delivery job, and pass its exact single-touch gate: independently approve the exact recipient_ref, channel, final rendered message/brief payload, and one concrete ISO-8601 dispatch_at plus timezone when scheduled, then run fresh eligibility and live-suppression checks immediately before the provider call. Any changed recipient, channel, payload, or schedule requires new approval.Disclosure and usage rights must be stated explicitly — never leave them as placeholders once the user has given input. Briefs are guidelines, not scripts: respect the creator's voice while pinning the key messages and compliance terms.
User: "Create a brief for our organic protein powder: 1 IG Reel + 1 TikTok, morning-routine/workout-fuel angles, approved clean-label claims, draft due 15 Sep, go-live 22 Sep, $1,200 fee, and 12-month repost/paid rights in the US."
Output: Complete reference-safe inline brief using the supplied claims, deliverables, dates, fee, territory, and rights scope, with dated official platform specs or TBD/NEEDS_INPUT. Offer the exact memory/influencer/brief-generator/YYYY-MM-DD-<topic>.md path; do not claim it was saved without exact authorization. Saving or a Send label never authorizes delivery; outreach resolves the raw render only after its separate exact send gate.
Termination note: Maintain a visited-set. If a recommended skill was already invoked this session, stop and report chain-complete instead of re-running it. Cap any handoff chain at max-depth 3.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 20,533 | 18,634 | -9% | 1 | 1 | 0% | 3,550 | 5,038 | +42% | 0 | 0 | — |
case-02 | fail→pass | 13,854 | 16,375 | +18% | 1 | 1 | 0% | 2,527 | 4,257 | +68% | 0 | 0 | — |
case-03 | fail→pass | 23,057 | 16,534 | -28% | 1 | 1 | 0% | 3,949 | 4,394 | +11% | 0 | 0 | — |
case-04 | fail→pass | 12,613 | 12,806 | +2% | 1 | 1 | 0% | 1,944 | 3,797 | +95% | 0 | 0 | — |
case-05 | fail→pass | 9,129 | 12,250 | +34% | 1 | 1 | 0% | 1,507 | 3,796 | +152% | 0 | 0 | — |
case-06 | fail→pass | 12,440 | 6,770 | -46% | 1 | 1 | 0% | 1,948 | 2,743 | +41% | 0 | 0 | — |
case-07 | fail→fail | 15,192 | 12,507 | -18% | 1 | 1 | 0% | 2,305 | 3,414 | +48% | 0 | 0 | — |
case-08 | pass→pass | 15,036 | 13,066 | -13% | 1 | 1 | 0% | 2,722 | 3,782 | +39% | 0 | 0 | — |
case-09 | pass→pass | 15,382 | 12,612 | -18% | 1 | 1 | 0% | 2,385 | 3,770 | +58% | 0 | 0 | — |
case-10 | pass→pass | 12,845 | 11,085 | -14% | 1 | 1 | 0% | 2,101 | 3,270 | +56% | 0 | 0 | — |
case-11 | pass→pass | 14,331 | 13,366 | -7% | 1 | 1 | 0% | 2,479 | 3,642 | +47% | 0 | 0 | — |
case-12 | pass→pass | 14,452 | 12,121 | -16% | 1 | 1 | 0% | 2,422 | 3,527 | +46% | 0 | 0 | — |
case-13 | fail→fail | 12,909 | 13,000 | +1% | 1 | 1 | 0% | 2,075 | 3,515 | +69% | 0 | 0 | — |
case-14 | fail→pass | 17,464 | 16,537 | -5% | 1 | 1 | 0% | 2,766 | 3,926 | +42% | 0 | 0 | — |
case-15 | fail→pass | 6,809 | 5,805 | -15% | 1 | 1 | 0% | 1,122 | 2,585 | +130% | 0 | 0 | — |
case-16 | fail→pass | 3,116 | 3,193 | +2% | 1 | 1 | 0% | 581 | 2,110 | +263% | 0 | 0 | — |
case-17 | fail→pass | 8,910 | 3,911 | -56% | 1 | 1 | 0% | 1,504 | 2,151 | +43% | 0 | 0 | — |
case-18 | fail→pass | 11,007 | 6,513 | -41% | 1 | 1 | 0% | 1,836 | 2,471 | +35% | 0 | 0 | — |
case-19 | fail→pass | 5,125 | 3,930 | -23% | 1 | 1 | 0% | 850 | 2,216 | +161% | 0 | 0 | — |
case-20 | fail→fail | 15,132 | 15,541 | +3% | 1 | 1 | 0% | 2,856 | 4,388 | +54% | 0 | 0 | — |
case-21 | fail→fail | 11,993 | 12,114 | +1% | 1 | 1 | 0% | 1,981 | 3,749 | +89% | 0 | 0 | — |
case-22 | fail→pass | 12,600 | 8,620 | -32% | 1 | 1 | 0% | 1,921 | 2,885 | +50% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +59 percentage points is the difference between those two pass rates over the 22 comparable cases.
The publisher has shipped newer versions since this run, so these numbers describe v1, not the version currently listed.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.