thimble plugin
Claude Code plugin that helps you verify agent work: visual answer cards, simplification and citation checks against your files.
- 116 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- A research prototype installed through `curl` that triggers extra Claude calls and reads your files.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/thimbleDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- A research prototype installed through `curl` that triggers extra Claude calls and reads your files.
This is third-party code. Review the repository files before installing.
What it does
thimble gives Claude Code tools to cite, visualize and label your data and an interface for understanding and verifying the agent's work faster. A key at an empty prompt starts an action: `v` builds an interactive card of the last answer, `s` restates it shorter and plainer, `c` asks a second Claude to check the selected words against your files, `e` explains the selection, `a` opens a side thread and `t` lists such threads. The plugin also adds explainers and heads-ups to answers on its own. There is also a browser workbench: the `thimble-browser` command in a folder of files. `/thimble-fix` asks Claude to change your own copy of the plugin and test the change, and `thimble file pr N` turns it into a pull request.
Who it is for. People who hand Claude Code long research and edits and want to grasp results faster and check them against sources.
Good fit when
- A long answer needs to be seen as an interactive card.
- A highlighted phrase should be checked against project files by a second Claude.
- A hard explanation has to be simplified without losing meaning.
- You want to park a clarifying question in a side thread without breaking the main conversation.
Not a fit when
- You need a guaranteed product; the author calls thimble a research prototype.
- The system is Windows without WSL 2; macOS and Linux are supported.
- Token spend is tightly limited; the second Claude for citations and the explainers use tokens.
Example request
Build a visual card of the last answer and check the highlighted figures against the files in this folder.Limitations
This is a research prototype and not an official Anthropic product, and it may have security vulnerabilities: the author asks that security issues be reported privately. macOS or Linux (WSL 2 on Windows), Claude Code, Python and Node.js are required. The install script runs from `curl` in the shell. Citation checks and added explainers trigger extra Claude calls and spend tokens and subscription limits. The plugin edits its own copy on `/thimble-fix`, and `thimble update` keeps such fixes. Install details and troubleshooting are in `INSTALL.md`.
How to disable. Turn features off by clicking `thimble` above the prompt. For full removal follow the section in `INSTALL.md` and delete the plugin through `/plugin`.
FAQ
How do I install thimble?
The recommended way: `claude "install thimble from https://github.com/safety-research/thimble"`. Manually: `curl -fsSL https://raw.githubusercontent.com/safety-research/thimble/main/scripts/install.sh | bash`.
What do the prompt keys do?
At an empty prompt `v` visualizes, `s` simplifies, `c` checks citations, `e` explains the selection, `a` opens a side thread and `t` lists threads.
How do I open the browser version?
Run `thimble-browser` in a folder of files. The terminal plugin runs with a plain `claude` in the same folder.
How do I change the plugin's behavior?
Type `/thimble-fix` and describe what should work differently. Claude changes your copy and tests it, and `thimble fixes` lists your fixes.
Which systems does it run on?
On macOS and Linux, and on Windows through WSL 2. Claude Code, Python and Node.js are required.
Discussion
No questions or reviews yet.
Related
PluginsA Claude Code plugin that makes the model finish the job: result verification, evidence and systematic investigation enforced as procedure
A self-improvement plugin for Claude Code, Codex and OpenCode: turns corrections into rules the agent follows in later sessions
A Claude Code plugin with hooks and 26 skills: measures what the agent knows, blocks edits until code is read, keeps memory
The author's personal Claude Code toolbox: meta-prompt and todo commands, skills to create skills, hooks and MCP servers, auditors and a Ralph loop
AI Builder Club Skills
AI Builder Club — Skills
A Claude Code plugin marketplace to make a repo agent-ready and run compounding agent loops
A plugin with six working skills and a prompt library: routing a task to a prompt, pre-commit checks and skill audits.
README in short
A plugin from an AI safety research group that simplifies checking Claude's answers through visualization, citation and side questions.