Skip to content

gooseworks-ai/goose-skills / skills/marketing/where-to-start

where-to-start

Verified

First-move advice for a broad growth or marketing ask with no job named: looks at what GooseWorks already knows and at the site, asks a few short questions in one message, names the one thing in the way, and sends the marketer to the right next job. Use when the marketer says "help me grow my brand", "I need to grow my business", "how do I get more customers / sales", "help me with marketing", "where do I start", "sales are flat", "what should I focus on this month", or "I'm back, what's next?". Recommends only; never runs the work. Not for a named job like "make me ads" or "grow my Instagram" (go straight to it); not for a campaign with a launch, budget or date — use plan-campaign; not for why this week's numbers moved — use analyze-performance.

FileKindSize
SKILL.mdskill md12.4 KB
references/base-rates.mdmarkdown5.1 KB
references/output-shape.mdmarkdown7.0 KB
references/playbook-map.mdmarkdown19.8 KB
references/questions.mdmarkdown5.1 KB
references/situation-lens.mdmarkdown9.7 KB
skill.meta.jsonconfig3.2 KB
tests/test-by-hand.mdmarkdown9.9 KB

tests/test-by-hand.md

View on GitHub
# Test by hand: where-to-start

## Setup
1. Connect GooseWorks **staging** to your agent.
2. Link every marketing skill side by side (the playbook needs its steps and the shared steps): see "How to test a playbook" in `docs/marketing/README.md`.
3. Open a fresh session for each case.

## Cases

Run at least the first five. Each uses a brand the reviewer knows first-hand where possible.

| # | Case | Say | Must happen | Fails if |
|---|---|---|---|---|
| 1 | Early brand, thin brain (ideally pre-launch or just launched) | "Help me with marketing." | Opens with what it knows; asks ≤ 5 plain questions (only gaps); names one constraint with evidence; recommends 1–2 moves (in plain words) in the output shape; offers a hand-off and stops. | Re-asks a known fact; recommends 3+; runs a playbook; reads more than three of the brand's own pages, or any competitor, ad-library or paid data; names a playbook or tool to the founder. |
| 2 | Stalled DTC brand with ads running (staging's Clearday; the reviewer plays the founder: about $40k/month revenue, $8k/month on Meta, Klaviyo used only for an occasional newsletter, two people, unsure tracking is right) | "Help me with marketing. Sales have been flat for two months." | Asks what changed when it flattened; names one constraint with evidence; uses the calendar (Black Friday is six weeks out on 16 Oct 2026); doesn't jump to new ads or more spend; OK-ing a few claim lines under "Bring" for any pick that writes copy (no Claims sign-off exists). | New ads or more spend as the first move; ignores the calendar; names a tool or platform the founder didn't (e.g. a store platform). |
| 3 | Fix-the-basics: a founder with no paying customers yet, or no tracking at all, who wants to run ads | "I want to start running ads, where do I start?" | Says plainly that ads are not the first move and why; the first move may be a human step (talk to customers, set up tracking) with at most one playbook alongside. | Recommends make-static-ads / run-paid-campaigns as the first move. |
| 4 | Off-map type: a snack brand selling mostly through 120 grocery stores, wanting more sell-through | "How do we sell more in stores?" | Names retail and trade marketing as off the map, gives the human first step (e.g. in-store demos, retailer promotions), and at most one playbook that helps; does not dress up an ad or outbound playbook as the answer. | Routes to the nearest playbook as if it covered retail. |
| 5 | Rich brain: a brand with signed-off claims on file and an earlier where-to-start visit | "What should I focus on this month?" | Asks 0–2 questions; reads back what it knows; treats it as a return visit (asks how the last recommendation went). | Asks the five questions from scratch. |
| 6 | Direct ask | "Make me ten ads for our new serum." | Hands straight to make-static-ads without the situation questions. | Runs the where-to-start interview. |
| 7 | B2B SaaS, five design partners, $0, founder sells, wants "LinkedIn and ads" | "We just launched, how do we get customers?" | Asks whether the partners use it weekly and would pay; constraint follows from that answer; picks are free or within credits the founder accepts; "not yet" covers ads with a reason; positioning-from-scratch gap named if relevant. | DTC logic (ads, photos) applied to B2B; a paid pick for a $0 founder without saying it costs credits. |
| 8 | Multi-brand organization, brand not named (staging org has 8 brands) | "Help me with marketing." | Asks which brand, by name, before any research; never defaults to one or silently picks the latest. | Diagnoses the wrong brand. |
| 9 | Return visit, didn't act (run case 2 first, then a month later) | "I didn't do any of it, too busy." | Opens from the saved block; doesn't re-ask why (they said); cuts the same move to the smallest slice before the next date. | Restarts the five questions; recommends something new just to be different. |
| 10 | Pre-launch app with an audience and no brand set up (habit tracker, TestFlight only, launch in three weeks, $2k, solo founder, 4k followers) | "Where do I start?" | Uses the existing audience before anything cold; respects the launch date; no paid installs before retention is known; says plainly the answers can't be saved until the brand is set up, and saves them if that happens in the same conversation. | Weeks of research while ignoring the 4k followers. |
| 11 | Earlier work saved on the brand by another session (e.g. a campaign plan with Meta ads and "claims not signed off"), founder with 1,000+ free sign-ups who mostly never come back, goal 20 paying by end of November, ~$1k/month, a few hours a week | "Help me with marketing." | Mentions the saved plan in one line with its date and asks whether it still stands; still asks where it's stuck, what they can put in, and the goal; names activation as the constraint; ads under "not yet"; a "worked if" with an outcome number and a date; says "saved" only after the save returns. | Builds on the saved plan unasked; names claims sign-off or account setup as the constraint; asks no situation questions; "worked if" is task completion; claims a save that didn't happen. |
| 12 | Agent-written brain (setup flags, a saved strategy, "claims not signed off"), one founding engineer, sign-ups who stall after trying it once; the founder says "people who left over price aren't our audience" but names solo founders as the best customers | "Where do we start with marketing?" | Treats the saved strategy and flags as proposals and checks them; ≤ 5 asks counting every sub-ask, including "any notes from user or cancel calls? paste them"; names activation, not acquisition, as the constraint; asks who the churners were before saving anything; product-scope calls shown as the founders' decision; moves sequenced to one person's hours; nothing saved before the founder agrees | Calls a setup step the constraint; asks 5 numbered questions holding 10+ asks; recommends social + cold email + creators at once; saves the buyer or strategy before a yes |

Also run at least one held-out case the reviewer invents on the day, unlike any example in `references/output-shape.md`, to check the answer comes from reasoning rather than the nearest example.

Round-2 additions to every case: the opening reads back what a short look at up to three of the brand's own pages showed (or says the host can't fetch pages); the recommendation names moves in plain words, never playbook or tool names; fails if it asks something the homepage answers.

## Scoring (the reviewer's rubric)

Pass a case only if every MUST is yes:
1. **Constraint (MUST):** names one binding constraint that matches the reviewer's own reading; the picks serve it.
2. **Stage (MUST):** the first move fits the stage; no buying attention before fit; no new traffic into a known leak.
3. **Specific (MUST):** cites at least two facts the founder gave; could not be pasted onto another company.
4. **Feasible (MUST):** every pick's inputs can be supplied this week within the stated budget and hours.
5. **Focused (MUST):** at most two playbooks, clear order, every idea the founder raised that wasn't picked appears under "not yet" with a reason.
6. **Measurable (MUST):** a "worked if" with a number and a timeframe.
7. **Honest (MUST):** no invented facts, no research done, nothing run, off-map human steps named as such, assumptions stated.
8. **Efficient (MUST):** ≤ 5 questions, ≤ 1 follow-up round, nothing known re-asked, answer at most 250 words.
9. **Leverage (nice):** uses an existing advantage (list, audience, network) and respects the next big date.
10. **Regret test (MUST):** would a senior advisor be embarrassed if the founder spent the next three weeks exactly here? If yes, fail.

Also check after each run: `brand/situation.md` (brand coworker scope) holds a new "Latest" entry, the previous one under History, and is findable as the brand document `Situation — <Brand> — <date>`, with the founder's words separate from the agent's read; the brand's onboarding profile holds only values the founder stated.

## Staging run log
See the bottom of this file (filled in by the builder session).

---

## Staging run log — 2026-10-16 (builder session, GooseWorks staging)

Case 2 setup (Clearday), steps 1–2 for real (all free reads); stopped before step 6's write (it would change the staging brand's onboarding profile and its situation file, which other sessions use).

**Step 1 reads:**
- `brand_read` sections summary, kit, products, onboarding, learnings, accounts (lean) → ~10 KB. Gives: sells acne-safe skincare (own site clearday.example), `brandType: beauty`, audience (women 20s–30s with acne-prone skin), 20 products with 3 tagged bestseller, voice, value props, tagline. Onboarding profile empty: role, ad spend, goals, platforms all missing. Accounts: none. Learnings: 1. `existingAds` is not in the lean kit (it appears with `detail: full`); own-ad presence also shows in summary as `ad_library_ad_count: 11` in the brand list.
- `account_whoami` → `user_context` null (first visit). (Round 2: the situation now lives in `brand/situation.md`; this read is no longer made.)
- `campaign_read` (list) → no campaigns.
- Claims check (regulated category): `knowledge_search ["claim"]` → 7 research-created "approved" claims and no "Claims sign-off" rule in learnings → OK-ing a few claim lines goes under "Bring" for any pick that writes copy.

**What the brain answered of the five questions:** Q1 fully (what, to whom, where: its own site). Q2 partly (ads running: 11 in the ad library; no revenue, no customer counts). Q3 not really (lean `messagingAngles` is empty; the ads/offers search returned nothing). Q4 and Q5 not at all. So the opening message reads back Q1, and asks Q2–Q5 (four questions).

**Would have run (writes, not executed):** `brand_update` `patch.onboarding` {role, ad_spend band, goals, platforms} as `user_correction` with the founder's words; the situation file `brand/situation.md` (brand coworker scope) through the `save-to-brain` skill (`../../save-to-brain/SKILL.md`) (round 2; earlier rounds used `user_context`, which is now dropped).
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 198kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 196kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 196kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 196kUpdated Claude CodeCodexVerifiedInstall