synthera--field-manual

The Slingshot
Method.

271 AI skills on board. We carry almost none of them, and reach for any in milliseconds. Same firepower, ~1% of the weight.

14,651tokens / turn — if stacked
161tokens / turn — actual
scroll · approach vector
live-demo

Try the Slingshot Pre-Router simulator.

Sample Prompts: "django tests async runner" "fix rendering lag on heavy web layout" "clean bigquery table partition"
Stage 1
L1 Keyword Match

Scanned 271 skill names in 4ms...

Stage 2
L2 Gemini via AGY

Intent matched skill ID: nextjs-performance

Stage 3
Context Injection

Loaded nextjs-performance.md into prompt hook (~140 tokens).

telemetry-calculator

Calculate your Token Drag Savings.

Total Installed Skills: 271 skills
Avg Messages per Session: 25 messages
362,250
Tokens Saved per Session

🚀 14.8x Session Range Increase

maneuver

Don't carry the cargo. Slingshot to it the instant you need it.

L1

Keyword pass

instant

The prompt is matched against 271 skill names in pure local code. "django tests" lands django-tdd in ~50ms. No model, no cost.

~0claude tokens
L2

The local brain

Gemini / AGY Subagent

When keywords miss — "make my page load faster" shares no word with react-performance — a local Ollama model or Gemini Flash via AGY reads the task and decides what's actually required.

0claude tokens
L3

Fetch on demand

pay-per-use

Only the one chosen skill's full playbook is read into context, only at the moment it's needed, then it's gone.

1 of 271loaded, briefly
build-it-yourself

Five moves. All local, all free.

$ curl -fsSL https://slingshot-method.pages.dev/install.sh | bash
01Put your skills in one folder. Each is a plain SKILL.md — a name, a description, instructions.
02Build a tiny index.json: every skill's name + description + path.
03Keyword search (find) — score the prompt against skill names. Instant, no model.
04Local router (route) — Ollama embeds or Gemini via AGY handles the misses. 0 API cost.
05A prompt hook wires it in: keyword first, brain on miss, surface the hit. Done.
Open the build manual