Tool Pick｜SpecStory CLI — Auto-Save Every Agent Coding Session as Searchable Markdown

🌏 中文版

Tool Info

What Problem It Solves

You spend an hour talking to Claude Code or Codex CLI, together tracking down a nasty bug and nailing down why some architectural decision has to be made a certain way. Close the terminal window, and that conversation only ever existed inside that one session — next time you hit the same problem, you re-explain the background from scratch. Want to share that debugging trail with a teammate? Screenshot it, or paste a wall of text with broken formatting and no way to search it later.

SpecStory CLI's approach is blunt and simple: it's a wrapper around the coding agent you already use. Swap your launch command from claude to specstory run claude, and the agent runs exactly the same — the only difference is that SpecStory saves every interaction alongside it as markdown, dropped into .specstory/history/ inside your project. Everything stays local by default; log in and you can optionally sync to SpecStory Cloud for full-text search and team sharing across machines and projects. Skip the login and you simply end up with a clean local record. It also ships a feature called Lore, which mines your saved session history into new agent skills — not written from memory, but distilled from commands you actually ran and that actually worked.

Good fit: developers who run Claude Code, Codex, Cursor CLI or similar tools from the terminal and often can't recall how they solved something, or who want to hand a complete debugging session to a teammate. If you only touch an agent occasionally and never look back at history, this won't do much for you.

Quick Start

Install

# macOS / Linux, via Homebrew
brew tap specstoryai/tap
brew install specstory

# Check which agents are already installed and supported
specstory check

Basic Usage

# Launch Claude Code through SpecStory; the session auto-saves
specstory run claude

# Same deal for Codex CLI or Cursor CLI — just swap the name
specstory run codex
specstory run cursor

# Re-render every past session in the project as markdown
specstory sync

After running this, your project gets a .specstory/history/ folder with one markdown file per session, ordered by time, ready to grep or feed into any full-text search tool.

Advanced Usage

# Optionally log in and sync local sessions to SpecStory Cloud
specstory login
specstory sync

# Install the Lore skill via npx, to mine past sessions into new skills
npx skills add specstoryai/getspecstory --skill lore

Once Lore is installed, run /lore in Claude Code ($lore in Codex, or just ask "mine my lore" in other agents) and it reads back through .specstory/history, distilling what you actually ran and what actually worked into an evidence-backed skill draft. Approve it, and it's installed into every supported agent on your machine.

Comparison with Existing Tools

Caveats

• The CLI is open source; the IDE extensions are not. The README is explicit about this — the Cursor and GitHub Copilot extensions are closed source. Only the terminal SpecStory CLI and the Lore skill have public source code.
• Cloud sync is opt-in, but remember the trigger. Stay logged out and everything stays local; once you run specstory login and then use specstory run, sessions auto-push to the cloud — worth knowing before you decide whether to log in.
• One more wrapper layer means one more potential failure point. It works by intercepting the target agent's input and output to save it. If a CLI interface changes in a future release, saving could break first — the project is open source, though, so you can send a PR for a new provider if it falls behind.

Today's Takeaway

Most discussion of "AI development memory" focuses on giving the agent long-term memory. SpecStory flips that around and solves the human's memory problem instead — a coding agent's conversation is itself a valuable engineering record, it's just that by default nobody saves it. Turning that record into searchable local files, then using Lore to distill "how we actually did it" back into a skill, lets a team's working practices grow organically out of real conversations instead of being written up by hand after the fact.

References

• specstoryai/getspecstory — GitHub
• SpecStory official site
• SpecStory full documentation
• Lore: distilling session history into agent skills