jev-pruner plugin
A Claude Code plugin that trims noisy Bash output with the Jev model before the result reaches the main model, without a summary
- 168 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Command output and context parts go to the external TypeSafe API and the plugin does not redact secrets
- The hook intercepts every Bash command result and trims it, so an important line can drop out of the reply
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/jev-prunerDetects 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.
- Command output and context parts go to the external TypeSafe API and the plugin does not redact secrets
- The hook intercepts every Bash command result and trims it, so an important line can drop out of the reply
This is third-party code. Review the repository files before installing.
What it does
jev-pruner sits as a hook between the Bash command and the main model response. If output exceeds 10,000 estimated tokens, the plugin splits it into chunks of 20 lines, and TypeSafe Jev answers one question per chunk: does any line in it need to stay. Needed chunks stay and the rest are replaced with removal markers. There is no summary, and the remaining lines go through as they are. Small output passes without a Jev call. Errors, JSON, XML, YAML, diff, binary data and whole-document commands such as `cat`, `jq`, `git diff` and `openssl` are left untouched. Source code, documentation and disassembly are also preserved. The full output is saved locally in `.claude/fast-jev-output/` before the first Jev request, and the markers point to that file, which Claude can read. The folder is gitignored by itself. Claude Code needs early-access function hooks, the variable `CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1` and a `TYPESAFE_API_KEY`. The plugin identifier is still `fast-jev-output`. The `minTokens`, `chunkLines` and `keepThreshold` options are set with `/plugin configure`. Codex has a separate build and skill.
Who it is for. Developers whose Claude Code sessions are clogged with long build and test logs and who have a TypeSafe key.
Good fit when
- A build prints tens of thousands of lines and only the needed chunks should reach the context
- A long session carries command output into every later turn and the token bill grows
- The full output is saved on disk and Claude reads it through a marker when needed
- The 10,000 token threshold is tuned through `minTokens` in the plugin settings
Not a fit when
- You have no TypeSafe key: without it pruning cannot call Jev and does not work
- Output contains secrets: the plugin does not redact them and sends content to Jev, and its check only stops archiving credential-like lines
- You need a lossless excerpt of `git diff` or JSON: the plugin keeps such outputs whole
Example request
Run the full project build and show how much output the plugin trimmed and where the saved full text livesLimitations
Claude Code with early-access function hooks, the variable `CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1` and a `TYPESAFE_API_KEY` are needed. The author warns that function hooks may change between Claude Code releases, and the declarations were checked on 2.1.274. Jev receives command output and context fragments through the TypeSafe API and its requests are billed. The plugin does not redact secrets, and only credential-like lines are excluded from the archive. Not every long output is trimmed: important parts stay whole. The plugin identifier is still `fast-jev-output` and not `jev-pruner`.
How to disable. Run `claude plugin uninstall fast-jev-output@fast-jev-output` and remove `CLAUDE_CODE_ENABLE_FUNCTION_HOOKS` and `TYPESAFE_API_KEY` from the settings. The `.claude/fast-jev-output/` folder can be deleted by hand.
FAQ
How do I install jev-pruner in Claude Code?
Enable function hooks and set the key in settings, then run `claude plugin marketplace add tamaratran/jev-pruner` and `claude plugin install fast-jev-output@fast-jev-output`. Start a new session.
When does the plugin leave output alone?
When output is 10,000 estimated tokens or less, on errors, JSON, XML, YAML, diff, binary data, whole-document commands and recognized source code or documentation.
Where is the full output kept?
In `.claude/fast-jev-output/`, and the removal markers reference the file. The folder gitignores itself. Credential-like lines are not archived.
What data goes to TypeSafe?
Command output and chunked conversation context go to the TypeSafe API so Jev can decide what to keep. The main model receives the already pruned result.
How is jev-pruner different from fast-jev-compaction?
Fast Jev Compaction removes unneeded tool calls from old history. jev-pruner works on a single Bash output right after the command, before it reaches the main model.
Discussion
No questions or reviews yet.
Related
PluginsTen Claude Code plugins: panes and bands above the prompt, secret handling, a vague-prompt filter and memory cleanup.
A marketplace of small Claude Code plugins: path previews, context and token meters, a smart /commit and turn rollback.
A plugin that readies a project for Claude Code: guidance, tests, review, TDD, refactoring and a resumable fix loop
A Claude Code plugin marketplace by Frad LEE: Git and GitFlow, GitHub, refactoring, memory, research, image generation and interface work
A plugin with a local MCP server: the agent scores its own code on a set of criteria through Jev and improves it by the numbers
Two Claude Code skills that look for an existing solution before coding and stop additions nobody asked for.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| jev-pruner (this card) | no | 168 | |
| fast-jev-compaction | no | 7,967 | removes unneeded tool calls from old history, while this trims fresh Bash output before it reaches the main model |
| RTK | no | 86,384 | a Rust proxy that condenses git and test output by rules, while here the Jev model decides on each chunk |
README in short
The README shows the chain from a Claude request to a Jev-trimmed reply and describes the threshold, chunking and Jev questions step by step. Command categories, information retention rules, Codex and Claude Code installation, saved output, a settings table, a data and privacy section and tests follow.