The Complete Production SOP: How to Deploy and Run Prompt Token Pruning & Redundancy Eliminator in Autonomous AI Agents

---
name: "prompt-token-pruning-compressor"
description: "Compress verbose contexts and conversation histories to cut token costs by up to 50%."
metadata:
  author: "Microsoft"
  version: "1.0.0"
  runtime_compatibility:
    - "codex"
    - "claude-code"
    - "cursor"
    - "windsurf"
triggers:
  - "compress prompt tokens"
  - "prune conversation history"
  - "reduce llm token cost"
permissions:
  fs_read: true
  fs_write: "workspace"
  shell_exec: false
  net_access: false
---

# Execution SOP & Rules
1. Strip repetitive conversational pleasantries, acknowledgment tokens, and boilerplate greetings.
2. Strip inline comments and formatting whitespace unless critical for algorithm explanation.
3. Execute extractive summarization to distill historical agent turns into essential variable state changes.
4. Protect and lock down critical entity names, URLs, schema definitions, and numeric values.

import fs from 'fs';

const content = fs.readFileSync(process.argv[2] || 0, 'utf-8');
const estimatedTokens = Math.ceil(content.length / 4);
console.log(`Estimated Token Footprint: ${estimatedTokens} tokens (${content.length} characters)`);

Prune conversation history and compress prompt tokens across our last 5 refactor cycles.
Preserve: function signatures, export contracts, and modified file paths.
Strip: verbose test execution stdout, standard boilerplate, and historical conversational pleasantries.

#!/usr/bin/env bash
set -euo pipefail
echo "Validating workspace syntax post-compression..."
pnpm tsc --noEmit
pnpm eslint . --max-warnings=0
echo "[SUCCESS] All syntax and entity contracts preserved perfectly."