LABSCO SUMMARY

It hands your model a structure to follow; the model still does the work.

The engine returns a prepared prompt plus the phase instructions for the selected framework, and the calling model carries those phases out itself — read that way this is a library and a chain runner rather than an extra reasoning layer, which is the right expectation to have before installing it. The gates are where the value concentrates: a criterion with a retry configuration, or a shell command that has to pass, turns "and then check it" into part of the chain. Whether it feels native or mediated depends on your client, since the shorthand operators only work where the plugin's hooks are installed.

What it is

A prompt-library server with three tools: an engine that runs stored prompts and chains, one resource tool that is CRUD over the prompts, gates and frameworks themselves, and a control tool for switching frameworks and reading state.

What you get

Stored prompts invoked by id with named arguments, versioned in YAML with hot reload and rollback, and a starting library of 39 prompts across 8 categories
Chains where one step feeds the next, resumable by id or restarted outright, with validation running between steps rather than only at the end
Gates as a first-class primitive: a criterion the model checks its own output against, blocking or advisory, and able to verify by running a shell command such as a test suite
Verification loops with try-and-timeout presets, so "implement, run the tests, read the failures, fix" is a configuration rather than a habit
Reasoning frameworks switched per run or persisted — CAGEERF, ReACT, 5W1H and SCAMPER, or one you author
Prompts, gates and frameworks created, updated, listed and rolled back through a single resource tool, with the model authoring new ones as it works
Templates exported as client-native skills, so a stored prompt becomes a slash command in your client with no MCP call at runtime
Server status, metrics and history for what is currently loaded

Requirements

Node and nothing to authenticate: no environment variables and no account. The bundled library gives you something to run on day one; your own resources live in a workspace the server scaffolds for you, or an existing one named by MCP_RESOURCES_PATH. Client plugins add hooks — so the shorthand operators work in conversation and chain state survives context compaction — on Claude Code and OpenCode, partly on Gemini CLI, and experimentally on Codex; any other MCP client gets the three tools without them. MIT.

Setup effort

One command — npx -y claude-prompts@latest --client claude-code

See installation options

USE CASES

01

Keeping a house prompt in one place instead of pasting it

Store the prompt once with its arguments declared, then invoke it by id in any session. Editing it later changes every future run rather than the copy in one conversation.

resource_managerprompt_engine

02

Picking a chain back up after it stopped halfway

A chain carries an id, so the next call resumes it rather than starting over, and a restart is an explicit choice instead of the default.

prompt_enginesystem_control

03

Putting a real check between two steps

Define the gate with its pass criteria and retry configuration, attach it to the chain, and answer it during the run with a verdict and an action — so a step that fails the criteria is retried rather than passed along.

resource_managerprompt_engine

04

Making the model keep going until the tests pass

Attach a shell verification to the step and let it iterate: implement, run, read the failure, fix, retry, within the try and timeout budget you set.

prompt_engineresource_manager

05

Trying a different reasoning frame on the same task

Switch the framework for the session or persist it, then run the same stored prompt to see whether the structure changes the answer.

system_controlprompt_engine

3 tools

Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.

`prompt_engine`

🚀 PROMPT ENGINE [CAGEERF]: Execute prompts with C.A.G.E.E.R.F framework and ground-truth validation.

WHAT IT RETURNS: Prepared prompt + CAGEERF phase instructions (Context→Analysis→Goals→Execution→Evaluation→Refinement). WHAT YOU DO: Execute each phase yourself using the returned structure.

SYNTAX:

>>prompt_id key:"value" # Run prompt with arguments
>>step1 --> >>step2 # Chain multiple steps
@CAGEERF >>prompt # Apply CAGEERF framework
:: verify:"npm test" :full # Shell validation (5 attempts, 5min)
:: verify:"cmd" loop:true # Autonomous until pass

PRESETS: :fast (1 try), :full (5 tries), :extended (10 tries) MODIFIERS: %clean (no injection), %lean (gates only), %judge (preview)

BEFORE YOU INSTALL

What to know first

Tools you get

3 tools when we connected.

Requires

npx on PATH

CONTEXT COST

Light Standing cost

Rated by the size of the 3 tool definitions that ride along on every turn, not by how many there are

INSTALLATION

Connect your client

Server setup

TERMINAL

npx -y claude-prompts@latest --client claude-code

MCP Clients

Claude Code

CLAUDE CODE · IN CLIENT

/plugin marketplace add minipuft/minipuft-plugins
/plugin install claude-prompts@minipuft

Claude Desktop

CLAUDE DESKTOP · JSON

{
  "mcpServers": {
    "claude-prompts": {
      "command": "npx",
      "args": [
        "-y",
        "claude-prompts@latest",
        "--client",
        "claude-code"
      ]
    }
  }
}

Codex

Codex hooks require Codex CLI 0.117 or later and are unavailable on Windows.

Enable hooks in ~/.codex/config.toml

CODEX · TOML

[features]
hooks = true

Then install the plugin

CODEX · TERMINAL

codex plugin marketplace add https://github.com/minipuft/minipuft-plugins.git
codex plugin add codex-prompts@minipuft

Restart Codex, run /hooks to review the plugin hooks, then try >>tech_evaluation_chain library:'zod' context:'API validation'.

OpenCode

OPENCODE · TERMINAL

npm install -g opencode-prompts
opencode-prompts install

Gemini CLI

GEMINI CLI · TERMINAL

gemini extensions install https://github.com/minipuft/gemini-prompts

Claude Prompts MCP Server

It hands your model a structure to follow; the model still does the work.

What it is

What you get

Requirements

Setup effort

Keeping a house prompt in one place instead of pasting it

Picking a chain back up after it stopped halfway

Putting a real check between two steps

Making the model keep going until the tests pass

Trying a different reasoning frame on the same task

3 tools

`prompt_engine`

`resource_manager`

`system_control`

What to know first

Tools you get

Requires

Light Standing cost

Connect your client

Server setup

MCP Clients

Claude Code

Claude Desktop

Codex

OpenCode

Gemini CLI

Claude Prompts MCP Server

It hands your model a structure to follow; the model still does the work.

What it is

What you get

Requirements

Setup effort

Keeping a house prompt in one place instead of pasting it

Picking a chain back up after it stopped halfway

Putting a real check between two steps

Making the model keep going until the tests pass

Trying a different reasoning frame on the same task

3 tools

prompt_engine

resource_manager

system_control

What to know first

Tools you get

Requires

Light Standing cost

Connect your client

Server setup

MCP Clients

Claude Code

Claude Desktop

Codex

OpenCode

Gemini CLI

Same category

@event4u/agent-config

Tripwire

LynxPrompt MCP

Promptheus

`prompt_engine`

`resource_manager`

`system_control`