williampsmith/marketing-skills

A Claude Code plugin for marketing done from inside the product repo, by an agent that can read
the code. It runs as a growth operator: it measures the funnel, picks the part that's holding growth
back, and works on it with skills that ship PRs and drafts for a human to approve. The deepest
track today is agent discoverability: getting a product found, read, cited and recommended by AI
answer engines (ChatGPT, Perplexity, Google AI Overviews and AI Mode, Gemini, Claude, Copilot) and
the agents built on them.
## Start here: gtm
/marketing:gtm is the single entry point. Run it once in a product repo (init), then let a
daily scheduler call it (tick).
### Tracks and focus: where the product is stuck
Growth work is split into tracks, each a ladder of rungs tested against data. gtm.py checks every
track on each tick and picks one focus: the track whose weakest funnel stage holds growth back,
filtered by what fits the product and who buys it.
mermaid
flowchart TD
P["GTM profile motion · deal size · cycle · where buyers look"] --> FIT["Fit which channels are worth trying"]
M["Metrics funnel counts · audits · access checks"]
subgraph T["Tracks (each a ladder of rungs)"]
direction LR
T0["Measurement instrumented → attributed → funnel visible"]
T1["Discoverability reachable → known → associated → preferred"]
T2["Demand indexed → ranking → visited → pipeline"]
T3["Conversion first step → activating → paying"]
T4["Retention retained → expanding → referring"]
T5["Paid (off until a human turns it on) tracked → profitable → scaling"]
end
M --> T
FIT --> F
T --> F["Focus rules 1 alerts first · 2 prelaunch: discoverability 3 measurement gates the funnel · 4 weakest stage 5 fit · 6 hold 30 days"]
F --> W["Work: 60% focus · 25% discoverability until preferred · rest explores"]
Discoverability is a standing priority: buyers in every sales motion now shortlist with AI
assistants, so it keeps a share of the work until it reaches preferred.
### What one scheduled run does
mermaid
flowchart TD
A["Daily scheduler (Hermes cron, /schedule, CI)"] --> B["gtm skill, tick mode"]
B --> C["gtm.py decides: focus, due jobs, budget, cooldowns, bets past their check date"]
C --> D["Run due jobs audit · access check · page watch · news watch · metrics sync outreach refresh · news stories worth acting on and, weekly, the work session: 2 actions, each a bet"]
D --> E["Metrics and audit scores written to the state store"]
D --> F
F["Outputs PRs for pages and site changes drafts for outreach, posts, pitches and profiles"]
D --> G["Digest in the state store, daily report with charts 'Needs you' items sent to chat (one each) or GitHub"]
F --> G
G --> H["Human merges PRs, sends drafts, approves with a reaction or reply, asks the operator questions in chat"]
E -.->|"bets close against the metrics"| C
Every improvement is a bet with a metric, a baseline, an expected value and a check date. When
the check date comes, gtm.py bet close judges it from the recorded metrics, and results feed back
into which tactics get tried next. A result becomes a learning only after a human accepts it.
| skill | what it does | treg spend |
| gtm | Entry point and operator: init (incl. pre-launch), daily tick, launch, status; focus, fit, bets, budget, digest | ~$8-12/month at defaults |