Jev Pruner — AI skill for Claude Code

jev-pruner

A Claude Code plugin that uses TypeSafe's Jev to trim noisy Bash output **after the command runs, but before its result is sent back to the main LLM**. This reduces the output carried into later turns without generating a summary.

Using Codex? See [Codex installation and usage](#codex).

Claude requests a Bash command → Command runs → Jev prunes stdout → Claude receives the result

1. A tool.call hook wraps the Bash tool's next() result.
2. Stdout of 10,000 estimated tokens or fewer passes through untouched, without reading history, writing an archive, or calling Jev. The gate uses estimateTokens on raw stdout, not a character count or an exact model tokenizer. minTokens can raise this threshold but cannot lower it. If Claude already saved the output to a file, the hook reads and counts that full output instead of its short preview. Errors, JSON/XML/YAML/diff/binary output, whole-document commands (cat, jq, git diff, git show, base64, and openssl) are left untouched.
3. Output is split into chunks of chunkLines lines, capped at 200 chunks; lines longer than 2,000 characters are split first.
4. Jev receives { context, task, history, command, chunks }, plus category and categoryGuidance for recognized build/test/install or search/excerpt commands, and one noul question per chunk: “does any line in this chunk need to remain available?”. A single needed line protects the chunk, including values required by earlier instructions even when the next reply must not repeat them. history includes user/assistant text, complete tool inputs, and tool-result text and structured data from the current session. Identical results attached to both a tool call and a result message are sent once, linked by their tool-use ID. Questions are batched so each request stays under 30,000 estimated tokens.
5. History and output share maxStateTokens, using a digit-aware estimate. H