PACK — Codebase Snapshot

Role

Model: Haiku (claude-haiku-4-5). Haiku is sufficient for this task. Sonnet or Opus will also work.

Pack the current project into three targeted Repomix snapshots via CLI:

Variant	File	Purpose
code	repomix-code.xml	Source code only — used by /cleanup, /frontend-audit, /backend-audit, /security-review
docs	repomix-docs.xml	Documentation only — used by /doc-audit
full	repomix-full.xml	Full codebase — fallback and ad-hoc use

Metadata stored at .pipeline/repomix-pack.json. All /qa audit agents read from their mapped snapshot variant instead of independently discovering files.

Hard Rules

1. Repomix must be installed. If repomix is not found on PATH, stop: "PACK BLOCKED — repomix is not installed. Run npm install -g repomix first."
2. Never modify source files. This skill only reads the codebase and writes to .pipeline/.

Process

Step 1: Resolve path

If an argument is provided, use it as the target directory. Otherwise use the current working directory.

Step 2: Check repomix is available

Run: command -v repomix

If not found, stop with the message from Hard Rule #1.

Step 3: Generate three snapshots

Run all three sequentially:

mkdir -p .pipeline

# Code snapshot — source code only, excludes docs/config/assets
repomix --compress --remove-empty-lines --no-file-summary --include-diffs \
  --ignore "**/*.md,**/*.mdx,**/*.rst,**/*.txt,docs/**,doc/**,*.config.*,*.json,*.yaml,*.yml,*.toml,*.lock,*.svg,*.png,*.jpg,*.gif,*.ico" \
  --output .pipeline/repomix-code.xml <resolved-path>

# Docs snapshot — documentation files only
repomix --remove-empty-lines --no-file-summary --no-directory-structure \
  --include "**/*.md,**/*.mdx,**/*.rst,**/*.txt,docs/**,doc/**,README*,CHANGELOG*,CONTRIBUTING*,LICENSE*" \
  --output .pipeline/repomix-docs.xml <resolved-path>

# Full snapshot — entire codebase
repomix --compress --remove-empty-lines \
  --output .pipeline/repomix-full.xml <resolved-path>

Step 4: Write state file

Write .pipeline/repomix-pack.json:

{
  "packedAt": "<current ISO timestamp>",
  "source": "<absolute path to packed directory>",
  "snapshots": {
    "code": {
      "filePath": "<absolute path to .pipeline/repomix-code.xml>",
      "fileSize": <bytes>
    },
    "docs": {
      "filePath": "<absolute path to .pipeline/repomix-docs.xml>",
      "fileSize": <bytes>
    },
    "full": {
      "filePath": "<absolute path to .pipeline/repomix-full.xml>",
      "fileSize": <bytes>
    }
  }
}

Get file sizes via Bash (wc -c < file or stat).

Step 5: Report

Report to user:

Pack complete.
  code:  .pipeline/repomix-code.xml  (<size>KB)
  docs:  .pipeline/repomix-docs.xml  (<size>KB)
  full:  .pipeline/repomix-full.xml  (<size>KB)
  Source: <path>

QA agents will use their mapped snapshot variant.
Run /qa to use these packs across all audits.

Output

Three snapshots written to .pipeline/. .pipeline/repomix-pack.json updated with paths, sizes, and timestamp. Run /qa to use these packs across all five audit agents.