Onchain Analyzer

An AI and prompt engineering assistant CLI. Despite the name, this tool is focused on helping you craft, optimize, and evaluate prompts for large language models. It provides commands for generating prompts, building prompt chains, comparing AI models, estimating token costs, and following safety guidelines.

All operations are logged with timestamps for auditing and stored locally in flat files.

Commands

Data Storage

All data is stored in the directory defined by the ONCHAIN_ANALYZER_DIR environment variable. If not set, it defaults to ~/.local/share/onchain-analyzer/.

Files created in the data directory:

• data.log — Main data log file (currently unused but created on init)
• history.log — Audit trail of every command executed with timestamps

Requirements

• bash 4.0 or later (uses set -euo pipefail)
• python3 — used by the cost command for token cost calculation (standard library only, no pip packages)
• Standard POSIX utilities — date, cat, echo, mkdir
• No external API keys or network access required

When to Use

1. Crafting prompts for LLMs — Use prompt and system to quickly scaffold well-structured prompts with role assignments and task definitions
2. Learning prompt engineering patterns — Use template to see common patterns (zero-shot, few-shot, chain-of-thought, role-play) and chain for multi-step reasoning workflows
3. Estimating API costs — Use cost to calculate approximate spend before sending large batches of tokens to an API
4. Comparing AI models — Use compare to get a quick reference of how GPT-4, Claude, and Gemini stack up in benchmarks
5. Ensuring responsible AI use — Use safety to review guardrails before deploying prompts in production environments

Examples

# Generate a prompt for a data analyst role
onchain-analyzer prompt "data analyst" "summarize sales data" "markdown table"
#=> Role: data analyst
#=>   Task: summarize sales data
#=>   Format: markdown table

# Create a system prompt for an expert role
onchain-analyzer system "cybersecurity researcher"
#=> You are an expert cybersecurity researcher. Be precise, helpful, and concise.

# View prompt chain methodology
onchain-analyzer chain
#=> Step 1: Understand | Step 2: Plan | Step 3: Execute | Step 4: Verify

# Estimate cost for 5000 tokens
onchain-analyzer cost 5000
#=> Tokens: ~5000 | Cost: ~$0.1500

# List available prompt templates
onchain-analyzer template
#=> 1. Zero-shot | 2. Few-shot | 3. Chain-of-thought | 4. Role-play

Configuration

Set the ONCHAIN_ANALYZER_DIR environment variable to change the data directory:

export ONCHAIN_ANALYZER_DIR="/path/to/custom/dir"

If unset, the tool respects XDG_DATA_HOME (defaulting to ~/.local/share/onchain-analyzer/).

How It Works

1. On every invocation, the tool ensures the data directory exists (mkdir -p)
2. The first argument selects the command via a case dispatch
3. Each command performs its action and appends an entry to history.log for auditing
4. The cost command uses an inline Python snippet to compute tokens × $0.00003
5. All output goes to stdout for easy piping and redirection

Powered by BytesAgain | bytesagain.com | [email protected]