repomix

name	repomix
description	Pack a repository or file subset into a single AI-friendly file using repomix. Load when preparing an external codebase for analysis, review, or planning — before scout, research, or peer-review workflows on large external repos.
updated	"2026-05-07T00:00:00.000Z"
tags	[]
related	["codebases","scm"]

repomix packing authority for codebase analysis preparation.

> Pack repository or file subset into single AI-friendly file. Use when target repo is too large for file-by-file loading. Grep before reading. Use `/tmp` for remote repos. For codebase context before packing, load `codebases` skill. For repository detection and metadata, load `scm` skill.

1. **Pack** — run `repomix` with appropriate flags. 2. **Check output** — note metrics: files processed, total tokens, output path. 3. **Grep before reading** — search output for patterns first. 4. **Read selectively** — offset/limit; avoid loading whole file. 5. **Clean up** — remove large output files when done; ephemeral artifacts.

<core_usage>

Pack current repo

repomix                          # Default: repomix-output.xml
repomix --style markdown         # Markdown
repomix --style plain            # Plain text

Pack specific dir/files

repomix path/to/dir
repomix --include "src/**/*.ts,**/*.md"
repomix --ignore "**/*.log,tmp/,dist/"

Remote repos — ALWAYS output to /tmp

repomix --remote user/repo -o /tmp/repo-analysis.xml
repomix --remote https://github.com/user/repo -o /tmp/repo-analysis.xml
repomix --remote user/repo --remote-branch main -o /tmp/repo-analysis.xml
repomix --remote user/repo --remote-branch 935b695 -o /tmp/repo-analysis.xml

Stdin file list

git ls-files "*.ts" | repomix --stdin
rg -l "TODO|FIXME" --type ts | repomix --stdin
git diff --name-only HEAD~5 | repomix --stdin   # changed files only

Clipboard

repomix --copy
repomix --include "src/**/*.ts" --copy

</core_usage>

Flag	Purpose
--style xml|markdown|plain|json	Output format (default xml)
--include "glob,glob"	Restrict to matches
--ignore "glob,glob"	Exclude matches
--compress	~70% token reduction (Tree-sitter) — repos >100k lines
--remove-comments	Strip comments
--output-show-line-numbers	Add line numbers
--include-diffs	Append working-tree + staged diffs
--include-logs	Append last 50 commits
--include-logs-count N	N commits instead of 50
--stdin	File list from stdin
--split-output SIZE	Split into files (e.g. 1mb, 500kb)
--token-count-tree	Token distribution tree, exit
--no-security-check	Skip secret scanning (caution)
--copy	To clipboard
-o FILE	Custom output
--init	Generate repomix.config.json

<token_analysis>

Before packing large repo, check distribution:

repomix --token-count-tree
repomix --token-count-tree 1000   # Nodes ≥1000 tokens

USE to identify token-heavy directories, decide --ignore, decide --compress.

</token_analysis>

<grep_patterns>

Search output rather than reading full:

OUTPUT=repomix-output.xml         # local default
# OUTPUT=/tmp/repo-analysis.xml   # remote

# Exports / public API
rg "export.*function|export.*class|export default" $OUTPUT
# Imports / dependencies
rg "import.*from|require\(" $OUTPUT
# Auth
rg -i "auth|login|password|token|jwt" $OUTPUT
# Endpoints / routing
rg -i "router\.|route\.|endpoint|\.get\(|\.post\(|\.put\(|\.delete\(" $OUTPUT
# DB / models
rg -i "model|schema|migration|query|sequelize|prisma|mongoose" $OUTPUT
# Error handling
rg "try\s*{|catch\s*\(|throw new|\.catch\(" $OUTPUT
# Config
rg -i "config\.|Config|configuration|\.env" $OUTPUT
# Tests
rg "describe\(|it\(|test\(|beforeEach\(" $OUTPUT

</grep_patterns>

<config_file>

repomix --init          # Creates repomix.config.json

Sets default include/ignore, format, style. Commit if project-specific.

</config_file>

<common_workflows>

Full snapshot for AI analysis

repomix --style markdown --compress --include-logs -o /tmp/snapshot.md

Targeted review of changed files

# --include-diffs appends ALL working-tree+staged diffs (not scoped to stdin files)
git diff --name-only HEAD~5 | repomix --stdin --include-diffs --style markdown --copy

Remote library inspection

repomix --remote yamadashy/repomix --compress -o /tmp/repomix-analysis.xml

Pack external repo for brainstorm/plan

repomix path/to/external/repo/src --compress --style markdown -o /tmp/codebase.md

Then load /tmp/codebase.md as context. Faster than file-by-file when AI needs holistic view.

</common_workflows>

<error_handling>

Problem	Action
Remote command fails	Verify slug/URL; check network; try --remote-branch main explicitly
Output too large	Use --compress; narrow --include; rg before reading
Pattern not found	Try broader regex; verify file tree
Permission error on local	Verify path exists and readable

</error_handling>

- Grep before reading — search output first; read sections only when grep gives target. - `/tmp` for remote repos — NEVER write remote output to working directory; pollutes `git status`. - USE `--compress` for repos >100k lines — ~70% token reduction, minimal information loss. - USE `--stdin` when targeted file list ready (`rg`, `git ls-files`, `git diff`). - Default XML — clearest file boundaries; switch to `--style markdown` for chat paste. - `repomix.config.json` keeps flags DRY — `repomix --init` for repeat repos. - Review output before sharing — built-in security check NOT exhaustive. NEVER pack files with API keys, certificates, credentials. - Before `--no-security-check`: load `{{AGENT_DIR}}/reference/security-rules.md`. - Output files ephemeral — add `repomix-output.*` to `.gitignore`; clean `/tmp` artifacts.

<output_rules>Output in English. Preserve verbatim CLI commands, flags, and grep patterns.</output_rules>