Skip to content

Latest commit

 

History

History
96 lines (64 loc) · 2.95 KB

ipfs_validator_README.md

File metadata and controls

96 lines (64 loc) · 2.95 KB

IPFS Payload Validation Script

This script compares the payload stored on IPFS for a governance proposal with the actual payload on the blockchain. It automates the process of fetching, validating, and analyzing data using an LLM (Large Language Model) and other APIs.

Features

  1. Blockchain Integration:

    • Fetches blockchain data via the ChainAPI for a specified chain.

  2. IPFS Data Retrieval:

    • Retrieves IPFS data for governance proposals using a scraping approach (temporary).

  3. Prompt Customization:

    • Supports custom templates for the LLM to validate and compare data.

  4. LLM Integration:

    • Sends prepared prompts to Anthropic's Claude AI for detailed analysis.

  5. Caching:

    • Implements caching for IPFS data to optimize performance and minimize redundant requests.

Requirements

External APIs

Currently only supports Claude.

  • Anthropic Claude: Requires an API key configured as an env variable ANTHROPIC_API_KEY. The model name configured as an env variable ANTHROPIC_MODEL, default model is claude-3-5-haiku-20241022.

Usage

Run the script using the command-line interface:

usage: ipfs_validator.py [-h] [--proposal_id PROPOSAL_ID] 
                         [--chain {ETH,ARB,AVAX,BASE,BSC,GNO,MET,OPT,POLY,SCROLL,ZKSYNC}]
                         [--proposal_address PROPOSAL_ADDRESS] 
                         [--prompt_templates PROMPT_TEMPLATES [PROMPT_TEMPLATES ...]]

Compare ipfs with actual payload.

options:
  -h, --help            show this help message and exit
  --proposal_id PROPOSAL_ID
                        The id of the proposal.
  --chain {ETH,ARB,AVAX,BASE,BSC,GNO,MET,OPT,POLY,SCROLL,ZKSYNC}
                        Blockchain chain.
  --proposal_address PROPOSAL_ADDRESS
                        Ethereum proposal address.
  --prompt_templates PROMPT_TEMPLATES [PROMPT_TEMPLATES ...]
                        Jinja templates for prompting the LLM.

for example:

python3 ipfs_validator.py --proposal_id 20 --chain SCROLL --proposal_address 0x2B25cb729D90630395Cd3140f3460a73A41Fe5f0

Workflow

  1. Parse Input Arguments: Validates and processes the provided command-line arguments.

  2. Fetch Blockchain Data: Retrieves source code for the specified proposal address using the blockchain's explorer API.

  3. Fetch IPFS Data: Extracts the IPFS hash for the proposal and fetches the corresponding data.

  4. Prepare Prompts: Constructs prompts using a predefined template, replacing placeholders with actual data.

  5. Analyze Data with LLM: Sends the prompts to Claude AI for analysis and prints the response.

Known Limitations

  • Fragile Scraping: The script currently scrapes Aave's governance UI for IPFS hashes. This approach is not reliable and should be replaced with an API when available.

  • Dependency on External APIs: The script requires access to Anthropic's Claude API and Aave's governance UI.