WEIFENG2333/phistory

Phistory automatically archives versioned system prompt snapshots from agent CLIs like Claude Code, Codex, OpenClaw, and Hermes.

What it solves

Phistory provides a historical archive and comparison tool for system prompts used by popular coding-agent CLIs. It allows developers and researchers to track how agent behavior, tool use, and runtime instructions change over time across different versions of these tools.

How it works

The project installs the exact CLI package for a supported agent and uses claude-tap to capture the HTTP request containing the prompt without actually calling the model provider. It also extracts static prompt strings from packages. These captures are stored as snapshots of prompts, traces, and metadata. GitHub Actions automatically check for new releases every hour to commit new snapshots.

Who it’s for

  • Agent builders looking to see how competitors iterate on prompt design.
  • Researchers needing stable, citable prompt snapshots for audits or reports.
  • Researchers and developers debugging agent behavior changes.

Highlights

  • Multi-agent support: Tracks prompts for a wide range of CLIs including Claude Code, Codex, DeepSeek Harness, and Grok Build.
  • Automated tracking: Hourly updates via GitHub Actions to capture new releases.
  • Prompt diffing: A web viewer allows users to compare snapshots across versions.
  • Static extraction: Ability to extract prompt-like strings directly from installed packages.

Related

  • Project
    anthropics/claude-quickstarts

    A collection of reference implementations and starter projects for developers to build AI agents, browser automation, and computer-use applications using the Claude API.

  • Project
    marckrenn/claude-code-changelog

    A community-maintained archive that tracks the evolution of prompts and feature flags for Claude Code across releases.

  • Project
    Piebald-AI/claude-code-system-prompts

    A comprehensive collection of extracted system prompts from Claude Code, providing transparency into the instructions used by its various agents and tools.

  • Project
    vercel-labs/skills

    A CLI for the open agent skills ecosystem that allows users to install and manage reusable instruction sets across various AI coding agents.

  • Project
    specstoryai/getspecstory

    A local-first knowledge management system that captures, indexes, and makes searchable all interactions with AI coding assistants across various IDEs and CLI tools.