cj-repo-init

name: CJ_repo-init description: "Detect which CJ_ skills are deployed, verify each one's per-repo prerequisites (cj-document-release.json, CJ-DOC-RELEASE.md, TODOS.md, work-items/ tree), print a health table, and on one confirm scaffold the missing repo-level prerequisites from generic portable seeds. Standalone utility — in-place, no worktree/ship. Use when: 'set up this repo for the CJ skills', 'init repo prerequisites', 'make this repo ready for CJ_', 'bootstrap repo config', 'verify repo prerequisites'." version: 0.1.0 allowed-tools: - Bash - Read - AskUserQuestion

Overview

/CJ_repo-init is a standalone utility that closes the seam between "the CJ_ skills are installed under ~/.claude/" and "the CJ_ skills actually work in this repo." A repo can have every skill deployed and still fail at runtime because a per-repo config file is missing:

• /CJ_document-release HALTs with [doc-sync-no-config] if cj-document-release.json is absent — and it runs at Step 5.5 of every cj_goal orchestrator, so a missing config breaks /CJ_goal_feature, /CJ_goal_defect, and /CJ_goal_todo_fix.
• CJ-DOC-RELEASE.md is the canonical prose contract for /CJ_document-release (the human/agent read that the machine config cj-document-release.json documents). It is checked for presence + a small required-headings set; needed by the same skills as cj-document-release.json.
• /CJ_suggest, /CJ_goal_todo_fix, and /CJ_improve-queue exit 1 if TODOS.md is missing.
• The scaffold → implement → qa pipeline expects a work-items/ tree.

This skill detects which CJ_ skills are deployed, maps each to its per-repo prerequisite(s), verifies the union, prints a health table, and (on one confirm) scaffolds the missing repo-level prerequisites from generic portable seeds. Idempotent: re-running on a healthy repo is a no-op that just prints the health table. In-place only — no worktree, no branch, no /ship.

Architecture (detection-in-script / AUQ-in-prose)

This skill follows the workbench's documented split (CLAUDE.md "Novel pattern callout", precedent scripts/skills-doc-sync-check): the detection + verification + scaffolding logic lives in a testable bash engine bundled with the skill (skills/CJ_repo-init/scripts/cj-repo-init.sh); this SKILL.md prose owns the single confirm AskUserQuestion. The script never prompts; the skill never re-implements detection.

Steps

Step 1: Resolve the engine path

The engine is bundled with the skill at skills/CJ_repo-init/scripts/cj-repo-init.sh, so it deploys alongside the skill into ~/.claude/skills/CJ_repo-init/scripts/ and needs no workbench root scripts/ or .source reach-back — that is what makes /CJ_repo-init genuinely standalone (it bootstraps a repo that has never seen the workbench). Resolve the repo-local copy when running inside the workbench, else the deployed skill copy:

_RI=""
_REPO_ROOT=$(git rev-parse --show-toplevel 2>/dev/null || echo "")
if [ -n "$_REPO_ROOT" ] && [ -x "$_REPO_ROOT/skills/CJ_repo-init/scripts/cj-repo-init.sh" ]; then
  _RI="$_REPO_ROOT/skills/CJ_repo-init/scripts/cj-repo-init.sh"
elif [ -x "$HOME/.claude/skills/CJ_repo-init/scripts/cj-repo-init.sh" ]; then
  _RI="$HOME/.claude/skills/CJ_repo-init/scripts/cj-repo-init.sh"
fi
[ -z "$_RI" ] && { echo "Error: cj-repo-init.sh not found. Run skills-deploy install."; exit 2; }
echo "ENGINE: $_RI"

If the engine is not found, tell the user to run skills-deploy install (or to run from inside the workbench) and stop.

If the engine reports "not inside a git repository" (exit 2), surface that message verbatim and stop — /CJ_repo-init operates on a repo's per-repo files.

Step 2: Run detection (always)

Run the engine in default mode and capture stdout + exit code:

OUT=$(bash "$_RI"); RC=$?
echo "$OUT"

Print the health table to the operator exactly as the engine emitted it (the prereq | needed-by | status table). Read the machine-readable tail:

• GAPS=<n> — number of repo-level gaps.
• INSTALL_GAPS=<n> — number of install-level gaps (reported, never auto-fixed by this skill).
• One REPO_GAP <prereq> <detail> line per repo-level gap.
• One INSTALL_GAP <prereq> <detail> line per install-level gap.

Step 3: Branch on the gap count

If GAPS=0: the repo is healthy at the repo level. Print a one-line confirmation ("Repo prerequisites: all present"). If INSTALL_GAPS>0, relay the INSTALL_GAP line(s) verbatim as advisory follow-up (they require skills-deploy install, which this skill never runs). Then stop — no AUQ, no writes. This is the idempotent no-op path.

If GAPS>0: surface exactly ONE AskUserQuestion. Quote the REPO_GAP lines so the operator sees precisely what will be created:

This repo is missing {N} CJ_ prerequisite(s): {one REPO_GAP line each}

Scaffold them now from generic portable seeds (creates cj-document-release.json, CJ-DOC-RELEASE.md, TODOS.md, and/or the work-items/ dirs — repo-level only; nothing under ~/.claude/ is touched)?

Options:

• Scaffold now (recommended) — runs --fix and re-prints the health table
• Dry-run only — re-print the report; make no changes
• Cancel — stop without changes

Install-level gaps (INSTALL_GAP) are NOT part of the scaffold offer — relay them as advisory text alongside the AUQ; they require skills-deploy install.

Step 4: Act on the choice

• Scaffold now: run bash "$_RI" --fix, print its full output (post-fix table + Scaffolded: list). The engine re-verifies internally, so the printed table is the authoritative post-fix state. Confirm to the operator which files were created. If any cj-document-release.json was present-but-invalid, the engine prints a NOTE: that it was NOT overwritten — relay that verbatim so the operator fixes it by hand.
• Dry-run only: run bash "$_RI" --dry-run and print its output. No writes.
• Cancel: print "Cancelled — no changes made." and stop.

Step 5: Done

This skill makes no commits and creates no branch. If the operator wants the scaffolded files committed, that is a separate /ship (or manual git) step they drive themselves. Re-running /CJ_repo-init on the now-healthy repo is a clean no-op (Step 3, GAPS=0).

Usage

/CJ_repo-init

No arguments. The single confirm AUQ is the only interaction, and only when repo-level gaps exist.

Error handling