SpecStory

SpecStory

by SpecStory (GitHub org specstoryai)

“Software should do what you meant.” Captures coding-agent conversations and turns them into searchable, shareable, reusable knowledge.

See https://specstory.com and https://github.com/specstoryai/getspecstory

Products (specstory.com, 2026-10-02)

• SpecStory (free): saves and syncs conversations and decisions from coding-agent sessions so prompts can be reused; free extensions, a CLI and cloud access; past conversations can be referenced with @ mentions.
• RunStory (private alpha): tests software changes against your codebase and feeds reproducible failures back to your coding agent.

Supported agents (GitHub README, 2026-10-02)

• IDE extensions: Cursor and GitHub Copilot in VS Code.
• Terminal agents (CLI, installed via brew tap specstoryai/tap && brew install specstory): Claude Code (Claude Code), Cursor CLI, Codex CLI (openai-codex-app), Droid CLI, DeepSeek TUI, Antigravity CLI, Grok Build, Muse Code, Qwen Code, OpenCode (opencode), Pi (pi-coding-agent) and Gemini CLI. Minimum versions documented in 2025 included Claude Code 1.0.27+, Codex CLI 0.42.0+ and Gemini CLI 0.15.1+ (docs.specstory.com, via search listing).
• The repo is Apache-2.0, about 1.3k stars and 89 forks; the CLI is open source and accepts new agent providers.

Features

• Local-first capture of conversations as git-friendly Markdown in a .specstory folder in the project; searchable history; @ references to earlier conversations in new sessions.
• Lore: mines saved sessions for what was actually run and what worked, and converts demonstrated workflows into installable agent skills (invoked with /lore in Claude Code or equivalent commands in other agents).
• Cloud sync and share links (share.specstory.com) for selected conversations; data stays local unless you share or enable sync.
• Earlier note claims not re-verified: “BearClaude” (a macOS spec-first app) and the ability to convert conversations to Cursor Rules / Copilot Instructions; treat as historical.

Use cases

Recovering context after resets or lost repos, attaching intent to pull requests for review, deriving rules and skills from past sessions, and auditing AI-assisted decisions. See also agents-md and the context-engineering notes for how such artifacts are consumed.

Limitations

Value depends on teams saving and curating sessions; captured files add repo noise unless you set a .gitignore policy; capture fidelity varies by agent integration. Cloud and team pricing is not published on the pages fetched, so it is not stated here.

Sources (accessed 2026-10-02)

• https://specstory.com
• https://github.com/specstoryai/getspecstory
• https://docs.specstory.com/integrations/terminal-coding-agents (from search listing; version minimums)