gooseworks-ai/goose-skills / skills/marketing/work-with-creators
work-with-creators
VerifiedReal creators and influencers, researched, priced and briefed. Use when the marketer says "find influencers", "which creators are worth paying, at what price?", "find real UGC creators to film for our ads", "who should we seed product to", "is this creator's engagement real", "what should I pay this creator" or "write a creator brief". Free research first; one credit yes only to vet deeper. Drafts only: contacts nobody, agrees nothing, moves no money. Not for AI-made UGC-style video — use goose-video; if unsure, ask: a real person filming, or AI-made?
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 13.0 KB |
| references/creator-brief.md | markdown | 9.0 KB |
| references/creators-by-company-type.md | markdown | 3.9 KB |
| references/data-sources.md | markdown | 5.9 KB |
| references/discovery.md | markdown | 5.2 KB |
| references/first-message.md | markdown | 2.9 KB |
| references/pricing-and-deals.md | markdown | 9.3 KB |
| references/roster-format.md | markdown | 9.0 KB |
| references/run-memory.md | markdown | 3.9 KB |
| references/vetting.md | markdown | 14.0 KB |
| references/wave-criteria.md | markdown | 3.4 KB |
| skill.meta.json | config | 4.8 KB |
| tests/test-by-hand.md | markdown | 6.8 KB |
tests/test-by-hand.md
View on GitHub# Test by hand — work-with-creators
## Setup
1. Connect GooseWorks **staging** to your agent and link this folder into your agent's skills folder (see the marketing-agent README).
2. Use a brand in a creator-heavy niche. On staging: **Clearday** (acne-safe skincare, 7 server-"approved" claims and no sign-off marker, 5 tracked competitors with organic feeds). A pet-supplements brand is the spec's example; any creator-heavy DTC brand works.
## The test
Ask: **"Which creators are worth paying to promote our acne serum, at what price?"** — with no budget or goal.
It passes if:
1. **No question before research.** The first thing you get is the report (the only allowed first question: "a real person filming, or AI-made?" when you say "UGC" without saying who films).
2. **Web research happened, for free.** The report names real creators found by web search and on public TikTok / Instagram / YouTube pages and rivals' collabs or partnership ads, with numbers read from real posts, dated and linked, labelled "public-page read" — never follower counts alone. What a page hid is listed, not guessed.
3. **One message, one credit yes.** After the report: three lines (best pick, first move, link), then only what changes the answer — budget (the report shows three wave sizes), gift unit cost if gifting, names to strike, claims lines to sign off, and **one** yes to "up to N credits" for a deeper vetting top-up that says what it adds.
4. **A document, not a chat summary.** `report.md` passes write-report §3 (its In short in §1's shape; Decisions for you holding the budget, names, claims lines and the top-up; at most six moves in What to do; What we're not doing; Sources with source skew; how sure as confirmed / likely / first look, never fact / inference; no file paths), has timely openings, the assumptions to correct, the shortlist with price range and math and **For you** per pick ("this creator's format fits your serum because…"), **You vs them**, wave 1 and its criteria, near-misses, "what we could not read" — and briefs plus first messages for the top picks. Offered to open (host with files) or shown as an artifact/document.
5. **Claims-safe.** No efficacy line for the serum unless signed off; lines needing proof say "you'd need proof of X first"; first messages carry no price and no claims.
6. **It contacts nobody**, and asks you to approve the names (all, or strike any) before anything is sent by you.
Then: give a budget and say "yes, up to 200 credits" → it re-prices, deepens vetting inside the ceiling (logs every charge), finishes the remaining briefs, re-delivers. You'd pay at least half the shortlist inside the stated range after spot-checking two picks.
Edge tests:
- Repeats: a monthly roster refresh is offered only when a programme is running or about to (with the reason); after a one-off shortlist or a single vet, no repeat is offered.
- "I need UGC for my ads" → asks "a real person filming, or AI-made?" first; AI-made goes elsewhere.
- B2B SaaS asking for LinkedIn creators and newsletters → LinkedIn and newsletter pages read in the browser, sponsor pages and media kits found, no gift-first, per-placement prices.
- On a re-run, paste fake wave-1 results → scored against `wave-1.md`, `wave_result` set, a roster change proposed for your yes.
## Staging run log
### 2026-10-16 — Clearday (0e5b2fb0…), up to the first paid step
What ran (all free):
- `brand_read` summary, accounts, organic, competitors → own accounts @clearday (IG), @cleardayskin (TikTok) with recent posts and numbers (e.g. IG reel 40,397 views / 1,462 likes / 31 comments); competitors Prequel Skin, CeraVe, CLEARSTEM, The Ordinary, Paula's Choice. **Competitor collabs visible for free** in the competitors' organic captions: The Ordinary × @recess_therapy (10th-anniversary series), Paula's Choice × @hannah_waddingham. Both are celebrity/large accounts — conflict notes for discovery, not picks for a micro wave.
- `knowledge_search` claims → 7 "approved" (server research approves them without review — the claims gate's contract treats them as candidates until a sign-off marker exists; `knowledge_search` learning "Claims sign-off" → empty, so no marker on Clearday yet): four are **press quotes** (Allure, Byrdie, NewBeauty), plus "Free Returns", "Free Shipping on Orders $45+", "Rated 4.9 out of 5 stars 234 Reviews". The kit's value props ("100% free of acne-triggering ingredients", "dermatologist-developed") are **context, not approved claims**.
- `knowledge_search` creators / disclosure / banned words → nothing on file: no creators in talks, no disclosure rules, no banned-word list beyond the kit voice "don'ts".
- `brand_read` kit, products, learnings → 20 products; the spray's record has **price $12.00, no unit cost or margin**; one learning (a video-script "don't").
- `brand_read` product record → the product description carries efficacy language ("bacteria-fighting", "daily defense against breakouts").
- `social_inspiration_library` → **empty** (research-market has not saved outliers yet).
- `campaign_read` → no campaigns. `file_list` query `work-with-creators` → no earlier roster.
What Gate 1 would ask and play back:
> Goal? (awareness, content for ads, or tracked sales) · budget for this wave? · platforms (TikTok + IG Reels suggested from your own channels) · creators already in talks? · unit cost of the spray and lip balm (gift cost) — not in the brand records · US only?
> Plan once answered: ~30 micro creators (acne / sensitive-skin routine niche) seeded with a routine bundle, gift-first (~6–12 posts expected) + 2–3 UGC-for-ads creators paid per asset; a credit ceiling for discovery (about 60–90 lookups; the first answer's charge shown before continuing).
Stopped here: the next call is paid.
First paid calls it would make (not run): `data_call_provider` scrapecreators `/v1/instagram/user/tagged-posts` on @clearday (existing fans first), `/v1/tiktok/search/keyword` "acne safe skincare routine", `/v1/facebook/adLibrary/company/ads` for CLEARSTEM and Prequel (partnership ads).
Findings from this run, fixed in the playbook:
- Workspace files are per user, not per brand → files go to the brand coworker workspace (the sibling convention).
- Server-"approved" claims include press quotes and a star rating → with no sign-off marker they are candidates and are cut from briefs; once signed off, press quotes are cited as the publication's words, not lines a creator says as their own experience; product descriptions are **not** a claims source — brief facts are neutral (format, size, key ingredients, how to use), never efficacy words, unless signed off.
- No unit cost in product records → the gift-vs-paid call must ask for it (already in "Before you start").
- Empty Inspiration library → discovery falls back to fans, competitor collabs and search; the playbook must not stall on it.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| add-product-e2e-evalpaperclipai/paperclipAdd or extend a Paperclip full-stack runner E2E workflow, fixture, matcher, or report evidence path for local or Daytona execution. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| add-runner-evalpaperclipai/paperclipAdd or extend a Paperclip Runner protocol evaluation definition, roster, assertion, or report fixture with provenance and narrow validation. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| diagnose-why-work-stoppedpaperclipai/paperclipCompatibility entry point for failed, stalled, looping, or over-recovered Paperclip work. Use fix-failed-paperclip-run for the full diagnosis, regression, repair, real verification, and green-PR workflow; honor an explicitly requested diagnosis-only scope. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| fix-failed-paperclip-runpaperclipai/paperclipDiagnose and fix failed, stuck, slow, or incorrectly blocked Paperclip tasks and runs. Use with a task/run URL or failure report to trace the root cause, prove a regression red then green, check second-order effects, verify the real workflow, and deliver a PR with passing checks. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |