setup

Setup

Interactive setup for metaswarm. Detects your stack, asks targeted questions, writes project-local files, and creates platform-appropriate instruction files and command shims. Replaces both npx metaswarm init and the old /metaswarm-setup command.

After Phase 2 (user questions), determine the correct coverage command from the detection results, then run this Bash command:

PLUGIN_ROOT="${PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT:-${extensionPath:-}}}"
if [ -z "$PLUGIN_ROOT" ]; then
  setup_script="$(find "${CODEX_HOME:-$HOME/.codex}/plugins/cache" -path '*/metaswarm/*/lib/setup-mandatory-files.sh' -print -quit 2>/dev/null)"
  if [ -n "$setup_script" ]; then
    PLUGIN_ROOT="$(cd "$(dirname "$setup_script")/.." && pwd)"
  fi
fi
if [ -z "$PLUGIN_ROOT" ] && [ -f "$(pwd)/lib/setup-mandatory-files.sh" ]; then
  PLUGIN_ROOT="$(pwd)"
fi
bash "${PLUGIN_ROOT}/lib/setup-mandatory-files.sh" "$(pwd)" <threshold> "<coverage-command>" --platform <platform>

Where:

• <threshold> is the user's chosen percentage (e.g., 100)
• <coverage-command> is the enforcement command for their test runner:
  • pytest → "pytest --cov --cov-fail-under=<threshold>"
  • vitest/pnpm → "pnpm vitest run --coverage"
  • jest/npm → "npx jest --coverage"
  • go → "go test -coverprofile=coverage.out ./..."
  • cargo → "cargo tarpaulin --fail-under <threshold>"
• <platform> is codex, claude, gemini, or all. Prefer:
  • codex when running in Codex (PLUGIN_ROOT or CODEX_HOME is present, or the user invoked $setup)
  • claude when running in Claude Code (CLAUDE_PLUGIN_ROOT is present, or the setup skill was invoked there)
  • gemini when running in Gemini (extensionPath is present, or the setup skill was invoked there)
  • all only when the user explicitly asks to configure every supported CLI

The script handles:

1. Instruction file — AGENTS.md for Codex, CLAUDE.md for Claude, GEMINI.md for Gemini; appends metaswarm section (or writes new), skips if already present
2. .coverage-thresholds.json — writes at project root with correct thresholds and command
3. Claude command shims — for Claude/all only, writes .claude/commands/start-task.md, prime.md, review-design.md, self-reflect.md, pr-shepherd.md, brainstorm.md

The script outputs JSON with what was created/skipped/errored. Check that "status": "ok".

If the script is not available or fails, fall back to writing these files manually with the Write tool. Do NOT skip them.