getspecstory

🐛 Bug Fixes

• Cursor IDE chats created in Cursor 3.27.14 or higher are saved again. That Cursor release changed how it stores a chat's message list, so SpecStory treated every new chat as empty: specstory sync, run and watch skipped it with "no conversation", and specstory list, search and resume left it out. Chats created before the Cursor update were not affected. Re-run specstory sync cursoride to save any chats that were skipped. Thanks to diegodtanton for the detailed report in issue 377.
• Cursor IDE chats from multi-root windows that were never saved as a .code-workspace file (folders added with "Add Folder to Workspace…") are now found. SpecStory only recognized saved .code-workspace files as multi-root workspaces, so specstory sync and watch run from one of the window's folders found none of its chats. Those chats are now saved to the first folder's .specstory/history, the same place a saved .code-workspace window already puts them. The same matching is used for VS Code Copilot. Re-run specstory sync cursoride from the first folder to save chats that were missed. Thanks to iemgnod for the report in issue 223.
• specstory list, search and resume now show Cursor IDE chats from a multi-root window under the window's first folder instead of under the .code-workspace file path, so the project they list is one you can run specstory sync from.

📢 Announcements

• The SpecStory CLI now supports Cursor Cloud Agents (i.e. cursorcloud), Cursor's hosted agents launched from https://cursor.com/agents, Slack, GitHub, Linear, the Cursor apps, or the Cloud Agents API, for sessions read from Cloud Agents API version 1.0.0. This provides the same support for saving to local markdown files and to the SpecStory Cloud as for Claude Code, Cursor CLI, Codex CLI, Factory's Droid CLI, Cursor IDE, Google's Antigravity CLI, DeepSeek TUI, VS Code Copilot, Muse Code, Pi, Qwen Code, Grok Build and OpenCode.
• Cursor Cloud agent sessions are never written to your local machine by Cursor, so SpecStory reads them from the Cursor API with your Cursor API key (provide via the CURSOR_API_KEY ENV var, or cursor_api_key under [providers] in ~/.specstory/cli/config.toml) and matches agents to your project by its git origin.
• The Cursor API serves full session transcripts, including tool calls and their output, for only 24 hours after a run finishes; after that it returns only the conversation text. SpecStory keeps a local copy of each Cursor Cloud Agent transcript in .specstory/cursorcloud/ in your project (or in the directory set by cursorcloud_mirror_dir in ~/.specstory/cli/config.toml), so a session synced within those 24 hours keeps its tool calls. Run specstory sync at least every 24 hours, or keep specstory watch running, to capture the details of each run in time. Don't delete the .specstory/cursorcloud/ directory in your project or a run older than 24 hours can only be rebuilt from the conversation text alone without its tool calls. The directory holds unredacted transcripts, so SpecStory adds a .gitignore file to that directory to keep it out of git.
• You can resume a Cursor Cloud Agent session into another agent, and resume any agent's session into a Cursor Cloud Agent. Resuming a Cursor Cloud Agent session into Cursor Cloud reopens the same agent. Resuming any other session starts a new cloud agent, which is a new paid run, seeded with the session so far. Sessions over 1,000,000 characters are too large to resume into a Cursor Cloud Agent.

🐛 Bug Fixes

• A session resumed with specstory resume into an agent that shows session titles (Cursor CLI, Cursor IDE, Cursor Cloud Agents, VS Code Copilot, Droid, DeepSeek TUI or OpenCode) is now titled with the original session's first prompt, as it is in specstory list and SpecStory Cloud. Previously it was titled with a short filename slug such as tell-me-all-about, and sessions from Claude Code, Codex CLI, Gemini CLI and Pi were titled "Resumed session".

🐛 Bug Fixes

• Your answers to Codex CLI's request_user_input questions are now included in saved markdown. Previously the question and options were saved, but the selected option and any note were dropped, leaving an empty result. Each question is now shown with its options and your answer underneath, and questions you cancelled are marked as unanswered. Thanks to daviddwlee84 for the detailed report and reproduction in issue 333.
• For Codex tools that SpecStory doesn't have a dedicated format for, the tool's arguments are now also shown as JSON. Previously they were lost whenever the tool also returned output.
• On Windows, specstory run and specstory watch no longer stop saving sessions partway through. Occasionally, when the agent created new session files while SpecStory was starting to watch a new directory, the file watcher could hang: no error was shown, and nothing more was saved to markdown or SpecStory Cloud until SpecStory was restarted. This could affect every supported agent.

📢 Announcements

• The SpecStory CLI now supports OpenCode (i.e. opencode), the open source terminal coding agent, for sessions created from OpenCode version 2.0.14 or higher. Sessions from earlier versions may work, but are not officially supported. This provides the same support for saving to local markdown files and to the SpecStory Cloud as for Claude Code, Cursor CLI, Codex CLI, Factory's Droid CLI, Cursor IDE, Google's Antigravity CLI, DeepSeek TUI, VS Code Copilot, Muse Code, Pi, Qwen Code and Grok Build.
• OpenCode is a cross-provider specstory resume target: you can resume any agent's session into OpenCode, and resume an OpenCode session into another agent. OpenCode's free Zen models decline to continue a conversation that was not produced through Zen, so a resumed session needs another configured model to continue. Thanks to marton78 for the original OpenCode provider contribution in PR 156.

✏️ Markdown Enhancements

• Grok Build's image, video, workflow, scheduler, and feedback tools now render as readable fields in saved markdown, with prompts and workflow scripts in code blocks, instead of as raw JSON.
• Grok Build's search_tool now lists each discovered tool's parameters (type, whether required, and description) instead of its full JSON schema. A single tool search can return hundreds of tools, so saved sessions that search connected integrations are now much shorter.
• Grok Build shell commands, monitors, background-task lookups, images, workflows, scheduled tasks, and feedback drafts now show their command, prompt, name, or ID in the collapsed tool header, so a session can be scanned without expanding each tool.
• Grok Build tool results no longer include Grok's notices about unrelated background tasks, or the wrapper tags around grep results. A background shell command's task ID, status, and output file now show as labeled fields.
• Grok Build use_tool results that are JSON are now shown in a JSON code block, and several tool fields now have consistently capitalized labels (e.g. URL, Query, Description).