@fortress-optimizer/openclaw-skill

OpenClaw skill for automatic prompt optimization via the Fortress Token Optimizer API. Reduces token usage by 15-30% without changing meaning.

Zero runtime dependencies. Uses native fetch (Node 22+).

Installation

# Via ClawHub
clawhub install fortress-optimizer

# Via npm
npm install @fortress-optimizer/openclaw-skill

Quick Start

import { registerSkill } from '@fortress-optimizer/openclaw-skill';

const skill = registerSkill({
  apiKey: process.env.FORTRESS_API_KEY,
});

// Use as OpenClaw context engine
const { contextEngine, hooks } = skill;

Standalone Client

import { FortressClient } from '@fortress-optimizer/openclaw-skill';

const client = new FortressClient({
  apiKey: 'fk_your_key_here',
  baseUrl: 'https://api.fortress-optimizer.com',
  level: 'balanced',
  provider: 'openai',
  timeout: 10_000,
  gracefulDegradation: true,
  minPromptLength: 50,
});

const result = await client.optimize('Your prompt text here...');
console.log(result.tokens.savings_percentage); // e.g. 23
console.log(result.optimization.optimized_prompt);

Configuration

Set via constructor options or environment variables:

| Option | Env Variable | Default | Description | |--------|-------------|---------|-------------| | apiKey | FORTRESS_API_KEY | (required) | Your API key from fortress-optimizer.com | | baseUrl | FORTRESS_BASE_URL | https://api.fortress-optimizer.com | API endpoint | | level | FORTRESS_LEVEL | balanced | conservative, balanced, or aggressive | | provider | FORTRESS_PROVIDER | openai | LLM provider hint | | timeout | - | 10000 | Request timeout in ms | | gracefulDegradation | - | true | Return original prompt on API failure | | minPromptLength | - | 50 | Skip optimization for short prompts |

Optimization Levels

• conservative - Light optimization, ~5% savings. Safe for all prompts.
• balanced - Deduplication + compression, ~15% savings. Good default.
• aggressive - Maximum optimization, ~30% savings. May rephrase.

Architecture

Your App
  |
  v
OpenClaw Skill (this package)
  |-- contextEngine  --> ingest/assemble flow
  |-- hooks          --> before-tool-call interception
  |-- client         --> HTTP calls to Fortress API
  |
  v
Fortress API (api.fortress-optimizer.com)
  |-- phrase compression
  |-- semantic deduplication
  |-- meta-removal
  |-- sentence optimization

Security

• HTTPS enforced (rejects non-HTTPS URLs except localhost)
• Response validation against prompt injection attacks
• Bearer token authentication
• No data stored client-side

API

registerSkill(config?)

Factory that returns { contextEngine, hooks } for OpenClaw integration.

FortressClient

• optimize(prompt) - Optimize a prompt, returns OptimizeResponse
• getUsage() - Get token usage stats, returns UsageResponse
• healthCheck() - Check API availability, returns boolean

FortressContextEngine

Implements the OpenClaw ContextEngine interface:

• bootstrap() - Health check (non-fatal on failure)
• ingest(content) - Queue content for optimization
• assemble() - Optimize and return result
• afterTurn(stats?) - Accumulate usage statistics
• getStats() - Get cumulative optimization stats

createBeforeToolCallHook(client, config)

Returns a hook that intercepts tool calls with prompt-like arguments and optimizes them before execution.

License

MIT