Chase AI+ Skills
Claude Code skills for the Chase AI+ community. This repo is a Claude Code plugin marketplace — install once, get every skill, update with one command.
Access to this repo comes through the Chase AI+ classroom. Keep the link inside the community.
Install
In Claude Code:
/plugin marketplace add cth9191/chase-ai-skills
/plugin install chase-ai@chase-ai-plus
To pull updates later:
/plugin marketplace update chase-ai-plus
Prefer a URL? You can point Claude Code straight at the repo instead of the
owner/repo shorthand — same result:
/plugin marketplace add https://github.com/cth9191/chase-ai-skills
…or just drop the repo URL into Claude Code and ask it to add the marketplace — it'll wire it up for you.
What's inside
loop-engineer
An interview-driven loop-engineering architect. You bring a task you want to automate; it does the hard part — the decisions, not just the code — then builds and wires the whole thing.
What happens when you run it
flowchart TD
A["You: describe a recurring task<br/>('every week I want Claude to…')"] --> B["1 · Diagnose<br/>where it sits on the ladder"]
B --> C["2 · Interview<br/>fast, one question at a time,<br/>each with a recommended answer"]
C --> D["3 · Build the full loop<br/>skill · trigger · state · verify · stop · README"]
D --> E["4 · Test run<br/>see one real output first"]
E --> F{Happy with it?}
F -->|yes| G["Wire it up<br/>cron · Claude routine · OS task · manual"]
G --> H["✅ Live & hands-off<br/>(you only show up at approval gates)"]
F -->|no| D
It also knows when a task shouldn't be a loop — a one-shot (use /goal) or
something where success can't be defined even by a human — and says so, instead of
building a token-burner.
The loop it builds for you
A loop is four phases plus a stop rule. The magic is the dotted line — each run reads what past runs learned and gets better:
flowchart LR
T["⏰ Trigger<br/>schedule / cron / event"] --> E["⚙️ Execution<br/>the skill — reads state FIRST"]
E --> V["✅ Verify<br/>the success criteria"]
V --> S[("🧠 State<br/>approach + score<br/>what worked / failed")]
S -. "next run learns from this" .-> E
V --> X{"🛑 Stop?<br/>goal hit · plateau · cap"}
SC["🔁 Scraper loop<br/>backfills lagged scores<br/>(e.g. likes/opens that<br/>arrive days later)"] -. "makes the score real" .-> S
Without State + the Scraper, a loop just repeats. With them, it improves — that's the whole point.
Success criteria — the 5-tier ladder it forces you to pick from
"If you get nothing else right, get this right." The skill makes you choose a real one:
| Tier | What it is | Example | Automatable? |
|---|---|---|---|
| 1 | Deterministic yes/no | tests pass, compiles | ✅ fully |
| 2 | Rule / constraint | "under 200ms", "no lint errors" | ✅ fully |
| 3 | A metric / number | runtime, open rate, views | ✅ (+ scraper if lagged) |
| 4 | Fuzzy → an LLM judge | "is this good writing?" | ⚠️ judge = a different model |
| 5 | Needs a human | taste, brand, high-stakes | 🙋 human approval gate |
The maturity ladder it diagnoses you on
flowchart LR
M["Manual<br/>(prove it by hand)"] --> K["Codified<br/>(a skill)"] --> A["Automated<br/>(on a trigger)"] --> L["Self-improving loop<br/>(state + criteria)"]
If your task is unproven, it still builds the whole loop — but bakes a validation gate into run #1: the loop produces one output and waits for your thumbs-up before it starts trusting its own scores.
Run it:
/chase-ai:loop-engineer
…or just describe a recurring task you want to automate and it triggers on its own.
See walkthrough.html for a full example session, start to finish.
© Chase AI. For use by Chase AI+ members. All rights reserved.