Skip to content

GEMINI.md

Latest commit

 

History

History
164 lines (133 loc) · 3.64 KB

GEMINI.md

File metadata and controls

164 lines (133 loc) · 3.64 KB

Gemini Prompts Extension

This extension provides reusable prompt templates, quality gates, and multi-step workflows for consistent AI outputs.

Quick Start

Invoke a Prompt

>>prompt_id

Example: >>deep_analysis, >>code_review, >>reasoning

Chain Prompts

>>analyze --> >>design --> >>implement

Chains execute sequentially with context passing between steps.

List Available Prompts

resource_manager(resource_type: "prompt", action: "list")

MCP Tools

prompt_engine

Execute prompts and chains with CAGEERF methodology.

Single prompt:

prompt_engine(command: ">>reasoning")

With arguments:

prompt_engine(command: ">>code_review", options: {"focus": "security"})

Chain execution:

prompt_engine(command: ">>analyze --> >>implement --> >>test")

Resume chain:

prompt_engine(chain_id: "chain-xyz#2", user_response: "...", gate_verdict: "GATE_REVIEW: PASS - ...")

resource_manager

CRUD operations for prompts, gates, and methodologies.

List resources:

resource_manager(resource_type: "prompt", action: "list")
resource_manager(resource_type: "gate", action: "list")
resource_manager(resource_type: "methodology", action: "list")

Create prompt:

resource_manager(
  resource_type: "prompt",
  action: "create",
  id: "my-prompt",
  name: "My Prompt",
  description: "Purpose",
  category: "development",
  system_message: "You are an expert...",
  user_message_template: "Task: {input}"
)

system_control

System administration and status.

system_control(action: "status")
system_control(action: "framework", operation: "list")

CAGEERF Methodology

Prompts follow the C.A.G.E.E.R.F framework:

Phase Purpose
Context Gather domain knowledge
Analysis Break down the problem
Goals Define success criteria
Execution Implement the solution
Evaluation Validate against goals
Refinement Iterate based on feedback
Finalization Complete and document

Quality Gates

Gates validate output quality between chain steps.

Gate response format:

GATE_REVIEW: PASS - All criteria met
GATE_REVIEW: FAIL - Missing error handling

Severity levels: critical | high | medium | low

Syntax Reference

Syntax Purpose
>>prompt_id Invoke single prompt
>>a --> >>b Chain prompts
:: 'criteria' Inline gate
chain_id Resume chain

Prompt Categories

  • development - Code-related tasks
  • guidance - Methodology and reasoning
  • analysis - Deep investigation
  • workflows - Multi-step processes

Optional: Enable Hooks

Hooks provide automatic >>prompt syntax detection. Add to your ~/.gemini/settings.json:

{
  "hooks": {
    "BeforeAgent": [
      {
        "matcher": "*",
        "hooks": [{
          "name": "prompt-suggest",
          "type": "command",
          "command": "python3 ~/.gemini/extensions/gemini-prompts/.gemini/hooks/prompt-suggest.py"
        }]
      }
    ],
    "AfterTool": [
      {
        "matcher": "prompt_engine",
        "hooks": [{
          "name": "chain-tracker",
          "type": "command",
          "command": "python3 ~/.gemini/extensions/gemini-prompts/.gemini/hooks/post-prompt-engine.py"
        }]
      }
    ],
    "PreCompress": [
      {
        "matcher": "*",
        "hooks": [{
          "name": "pre-compact",
          "type": "command",
          "command": "python3 ~/.gemini/extensions/gemini-prompts/hooks/gemini/pre-compact.py"
        }]
      }
    ]
  }
}

Without hooks, use MCP tools directly: prompt_engine(command: ">>reasoning")