Inspect conductor.json, .worktreeinclude, existing conductor-setup.sh, conductor-run.sh, conductor-shutdown.sh, conductor-archive.sh, package scripts, and repo docs.
Read references/conductor-docs.md before changing script fields or environment variable usage.
Prefer committed conductor.json for shared team behavior. Remember that local Repository Settings can override it.
Keep setup, run, and archive/shutdown responsibilities separate:
- scripts.setup: prepare a new workspace.
- scripts.run: start long-running app/server/test processes.
- scripts.archive: run shutdown cleanup before a workspace is archived.
Use Conductor variables instead of hard-coded paths or ports:
- $CONDUCTOR_WORKSPACE_PATH for the active workspace.
- $CONDUCTOR_ROOT_PATH for shared root-level resources and caches.
- $CONDUCTOR_WORKSPACE_NAME for workspace-specific names.
- $CONDUCTOR_PORT for server ports.
For monorepos and multi-repo systems, prefer Conductor's working-directory selection, /add-dir, or per-repository run scripts over hard-coded sibling checkout paths.
Validate JSON syntax and run the narrowest relevant local check for any script you touch.

Required Conventions

Always use copy-on-write cloning for files over 100 MB when duplicating large artifacts. On macOS prefer cp -c; on Linux prefer cp --reflink=auto. If clone/reflink is unavailable, prefer a symlink or shared cache under $CONDUCTOR_ROOT_PATH unless the repo needs a true copy.
Route every conductor-setup.sh run to conductor-setup.log so setup failures are easy to debug in loops. Preserve exit codes with pipefail when piping through tee.
Use a committed .worktreeinclude for static gitignored files that should be copied into every workspace. Keep setup scripts for commands, generated files, symlinks, and workspace-specific resources.
Put required PATH/toolchain setup directly in setup or run scripts when possible. Do not rely on slow, interactive, or prompt-producing shell startup files.
Put shared caches outside individual workspaces under $CONDUCTOR_ROOT_PATH, and use the archive/shutdown script to flush, update, compact, or clean those shared caches.
Do not background long-running run-script processes with &. Use concurrently, a supervisor, or a single foreground process group so Conductor can stop everything cleanly.
Use runScriptMode: "nonconcurrent" only when the project cannot safely run multiple workspace instances because of a single fixed port, database, or shared local resource.
Use Spotlight testing instead of overcomplicated run/setup scripts when the project truly needs the repository root checkout, expensive root build artifacts, or one heavy local stack.
Do not commit secrets, provider API keys, or machine-specific credentials in conductor.json. Only change .mcp.json or enterpriseDataPrivacy when the user asks or the repository policy requires it.

Snippet

Use this as a starting point and adapt command names to the repository:

{
  "scripts": {
    "setup": "bash -lc 'set -o pipefail; ./conductor-setup.sh 2>&1 | tee conductor-setup.log'",
    "run": "./conductor-run.sh",
    "archive": "./conductor-shutdown.sh"
  },
  "runScriptMode": "concurrent"
}

Inside conductor-setup.sh, make large-file copies explicit:

copy_large_file() {
  src="$1"
  dest="$2"
  size="$(wc -c < "$src" | tr -d ' ')"
  if [ "$size" -gt 104857600 ]; then
    cp -c "$src" "$dest" 2>/dev/null || cp --reflink=auto "$src" "$dest" 2>/dev/null || ln -s "$src" "$dest"
  else
    cp "$src" "$dest"
  fi
}

Review Checklist

conductor.json is valid JSON and uses only documented fields.
Setup work that only copies static gitignored files is not over-engineered; prefer .worktreeinclude for team-shared file-copy patterns.
Scripts contain their own required shell setup and do not depend on interactive prompts or slow startup files.
Setup logs land in conductor-setup.log.
Run scripts use $CONDUCTOR_PORT or runScriptMode: "nonconcurrent" when needed.
Shared caches live under $CONDUCTOR_ROOT_PATH, not inside transient workspace directories.
Shutdown/archive logic updates shared caches and cleans only resources owned by the workspace.
Root-only, fixed-resource, or very expensive local stacks are handled with Spotlight testing or nonconcurrent, not brittle workspace path hacks.
Monorepo and multi-repository needs are handled with Conductor working-directory selection, /add-dir, or separate per-repository scripts.

conductor-setup

Conductor Setup

Workflow