Tools & skills index
Skills auto-invoke off their description frontmatter, but the measured honor rate on this machine
is ~4% — a design that assumes "Claude will pick the right skill automatically" fails 24 times
out of 25. So: invoke by name, deliberately. For every task below, say the exact skill name.
This index is also what makes the access ladder legible — a skill you don't see
listed at your current stage isn't unlocked yet. Ask your manager (Ollie) if you think that's wrong.
Access ladder (spec §04)
| Stage |
What you can do |
| 0 |
Manual only — no Claude, no tooling. Proves method. |
| 1 |
Read-only skills — research, audits, analysis. Nothing writes. |
| 2 |
Draft-generating — briefs, copy, LP drafts, campaign drafts. Output goes to review, never live. |
| 3 |
Build-but-PAUSED — real entities created in real accounts, always paused. |
| 4 |
Spend-touching — activate, budget change. Requires an explicit unlock from Ollie, per workstream. |
Day-one stage and ceiling per workstream (spec §04 table):
| Workstream |
Day one |
Ceiling |
| W-G Research |
2–3 |
3 |
| W-B Creative |
2 |
3 |
| W-C Copy |
2 |
3 |
| W-D Landing pages |
2 |
3 |
| W-F Organic social |
1–2 |
3 |
| W-H Store ops |
1 |
3 |
| W-E Email / SMS |
1 |
3 |
| W-A Meta ads |
0 |
3 → 4 late |
| W-K Tool health |
2 |
3 (gated by the regression gate, not stage — see K1–K4) |
W-A · Meta Ads
| Skill |
Does |
Needs as input |
Stage |
scale-gate |
Scale/kill threshold check against breakeven |
ad set ID, spend/ROAS window |
1 (read-only; the decision stays Ollie's) |
launch-doctor |
Pre-launch sanity check on a campaign build |
campaign draft |
1 |
meta-ad-launcher |
Builds a campaign/ad set/ad — PAUSED only |
brief, creative, copy, budget |
3 |
meta-ads-batch |
Launches a batch of creative-test ads into an existing ad set |
n creatives + fixed budget/duration |
3 |
meta-creative-abcd |
Structures an A/B/C/D creative test |
creative variants |
2 |
performance-marketer |
Full ROAS/CPA/kill-scale analysis — the deterministic decision engine |
brand, date range |
1 (reads standing verdicts; stays Ollie's call at stage < 4) |
ad-copywriter |
Meta ad-unit copy (Primary Text / Headline / Description) |
angle, brief or creative URL |
2 |
angle-researcher |
Angle generation from product/customer research |
product, competitor context |
1–2 |
W-B · Creative Production
| Skill |
Does |
Needs as input |
Stage |
kive-brief-generator |
Structured creative brief for the Kive pipeline |
product, angle, reference |
2 |
hf-product-ugc, hf-ugc, hf-static-ad, hf-packshot, hf-product-shot, hf-model-product-shoot, hf-clothing-shoot, hf-reformat, hf-add-motion, hf-multiply-ugc, hf-product-swap-ugc, hf-style, hf-model, hf-brand |
Higgsfield generation pipeline — ~20 skills, one per creative shape |
product images, references, brief |
2 |
ad-variant-factory |
Batch-generates creative variants from a working base |
base creative |
2 |
ad-creative-audit |
Pre-live creative QA against brand standards |
finished creative |
1–2 |
on-creative-copy |
On-creative overlay copy (Headline/Subtext) |
angle, creative |
2 |
nano-banana-prompts |
Prompt construction for the nano-banana image model |
brief |
2 |
boutique-creative-research / competitor-research |
Swipe / reference sourcing |
competitor URL or niche |
1 |
grab-media |
Downloads competitor creative from a store/product/Ad Library URL into the swipe folder |
competitor URL |
1 |
W-C · Copy
| Skill |
Does |
Needs as input |
Stage |
ad-copywriter |
Meta ad-unit copy — Mode 1 (pipeline) or Mode 2 (ad-hoc from a creative URL) |
angle ID or creative URL |
2 |
angle-researcher |
Angle → copy upstream |
product/customer research |
2 |
copy-editing / copy-pass |
Copy QA and editing pass |
draft copy |
1–2 |
on-creative-copy |
On-creative overlay text |
angle, creative |
2 |
W-D · Landing Pages
| Skill |
Does |
Needs as input |
Stage |
lp-factory |
End-to-end LP pipeline orchestrator — research → build → review → launch |
LP Factory Sheet row |
2 (publish gated to 3+) |
lp-copywriter, lp-angle-strategist, lp-hook-generator, lp-offer-strategist |
LP copy layer |
product, angle |
2 |
lp-builder / lp-shopify-builder |
Builds the LP |
brief + copy |
2–3 |
lp-qa-bot / lp-copy-qa-bot |
Pre-publish QA |
built LP |
1–2 |
lp-competitor-lp-analyzer / lp-ad-library-researcher / lp-market-researcher |
Research layer |
product/niche |
1 |
lp-audience-builder |
Audience definition for the LP |
product, angle |
1–2 |
W-E · Email / SMS
| Skill |
Does |
Needs as input |
Stage |
klaviyo-campaign-builder |
Builds a Klaviyo campaign — draft only |
brief, audience, offer |
2 |
klaviyo-lifecycle-flows |
Flow maintenance (welcome, abandon, post-purchase) |
flow spec |
2 |
klaviyo-email |
Actually sends. Stays Ollie's — never delegate below stage 4 with an explicit unlock. |
— |
4 only |
sms |
SMS campaign drafting |
brief |
2 |
W-F · Organic Social + Community
| Skill |
Does |
Needs as input |
Stage |
organic-social |
Content calendar + drafts — cannot publish (T2 gap, see spec §09) |
content plan |
1–2 |
(net-new, not yet built) social-inbox-manager |
Comment/DM triage across IG/FB/TikTok — T1 in spec §09, confirmed zero coverage today |
— |
not yet available |
referrals |
Creator/affiliate outreach |
— |
1–2 |
W-G · Research
| Skill |
Does |
Needs as input |
Stage |
product-research |
Product validation scoring |
product candidate |
1 |
competitor-research / competitor-tracker / track-competitor / brand-track |
Competitor sweep, tracker dashboard queries |
brand/domain |
1 |
customer-research |
Customer/review mining |
product |
1 |
pricing |
Pricing analysis |
product |
1 |
W-H · Store Ops
| Skill |
Does |
Needs as input |
Stage |
product-upload |
New DN product listing, Kive PDP image generation |
product data, images |
2–3 |
pdp-review-block |
PDP review-block content |
product |
1–2 |
shopify-plugin skills (shopify-admin, shopify-customer, shopify-liquid, etc.) |
Store admin, theme, customer ops |
varies |
1–3 per skill |
W-I · Reporting + Rhythm
| Skill |
Does |
Needs as input |
Stage |
analytics |
Margin/traffic analysis |
date range |
1 |
post-build-verification |
8-point silent-failure checklist after any build |
just-built artifact |
1 (self-run, mandatory) |
daily-brief |
Daily rollup |
— |
1 |
seo / ai-seo / seo-audit |
SEO checks |
page/site |
1 |
W-K · Tool Health
Not gated by stage — see spec §04 "Tool health doesn't use this ladder." Gated by whether the
regression suite goes green, not by tenure.
| Skill |
Does |
Needs as input |
system-triage |
Diagnoses a red system from the registry, proposes a fix (never applies silently) |
system id |
gate-doctor |
Reads the regression-gate state |
— |
cron-doctor |
Diagnoses a failing scheduled job |
job name |
mcp-doctor |
Diagnoses a broken MCP server |
server name |
Cross-cutting
| Skill |
Does |
When |
post-build-verification |
Silent-failure checklist |
after ANY build touching an API/DB/filter/pagination |
pre-deploy-security |
Security checklist |
before ANYTHING goes live on the web |
self-critique |
Adversarial review of a spec/plan/deliverable before presenting it |
non-trivial output, before asking for approval |
github-safety-check |
Vets third-party code before install |
before npm install / pip install / cloning a repo |