Skills

Skill 2389-research/speed-run/skills

Token-efficient code generation pipeline - parallel implementation with hosted LLM (Cerebras) for ~60% token savings. Includes MCP server.

Install

npx -y skills add 2389-research/speed-run --skill skills

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Token-efficient code generation pipeline using hosted LLM. Triggers on "speed-run", "fast build", "turbo build", "use hosted LLM", "use cerebras". Routes to turbo (direct codegen), showdown (parallel competition), or any% (parallel exploration).

SKILL.md

2.6 KB, as published. Nobody here has run it

Speed-Run

Token-efficient code generation pipeline. Uses hosted LLM (Cerebras) for fast, cheap first-pass generation. Claude handles architecture and surgical fixes.

Announce: "I'm using speed-run for token-efficient code generation."

Step 1: Check API Key (MANDATORY)

ALWAYS run this first:

mcp__speed-run__check_status

If status is error / key not set:

Speed-run requires a Cerebras API key for hosted code generation.

Setup (pick one):

Option A (recommended): Add to ~/.claude/settings.json:
  { "env": { "CEREBRAS_API_KEY": "your-key" } }

Option B: Add to ~/.zshrc:
  export CEREBRAS_API_KEY="your-key"

Get a free key at: https://cloud.cerebras.ai

Then restart Claude Code.

STOP here. Do not fall back to Claude direct generation — the whole point of speed-run is hosted LLM. If the user wants Claude direct, they should use test-kitchen instead.

Step 2: Route

If check_status returned OK, present options:

Speed-run ready! Cerebras API connected.

How would you like to proceed?

1. Turbo - Direct code generation (single task, fast)
   → Best for: one feature, algorithmic code, boilerplate
2. Showdown - Parallel competition (same design, multiple runners)
   → Best for: medium-high complexity, want best implementation
3. Any% - Parallel exploration (different approaches)
   → Best for: unsure of architecture, want to compare designs

Routing:

Option 1: Invoke speed-run:turbo
Option 2: Invoke speed-run:showdown
Option 3: Invoke speed-run:any-percent

Skill Dependencies

Subskill	Description
`speed-run:turbo`	Direct hosted codegen via contract prompts
`speed-run:showdown`	Parallel same-design competition via hosted LLM
`speed-run:any-percent`	Parallel approach exploration via hosted LLM

When to Use Speed-Run vs Test Kitchen

Situation	Use
Want token savings on code generation	Speed-run
Generating algorithmic code (parsers, state machines)	Speed-run:turbo
Want parallel competition with fast generation	Speed-run:showdown
Want to explore approaches with fast generation	Speed-run:any%
No API key / don't want external LLM	Test Kitchen
CRUD / simple operations	Test Kitchen (Claude direct is cheaper)
Need 100% first-pass accuracy	Test Kitchen