pakhale

My personal CLI. It keeps my coding agents — Claude Code and opencode — set up the same way on every machine, from one file.

npx pakhale setup agents          # set up this machine
npx pakhale setup agents plan     # dry run — show every change, write nothing

Running plan first is always safe: it reads the machine and prints exactly what would change.

What it sets up

config.ts is the whole source of truth. It lists the agents I use and the capabilities I want them to have — plugins, skills, MCP servers — plus my shared instructions and settings. Applying it installs whatever is missing and leaves everything else alone.

Agents package things differently, so each capability says how it reaches each agent:

{
  name: 'plannotator',
  deliver: {
    'claude-code': { via: 'claude-plugin', marketplace: 'plannotator', plugin: 'plannotator' },
    opencode: { via: 'opencode-plugin', package: '@plannotator/opencode@latest' },
  },
}

| via | Reaches | | --- | --- | | skills | any agent that reads SKILL.md | | mcp | any agent | | claude-plugin | Claude Code | | opencode-plugin | opencode |

Leaving an agent undeclared is an error rather than a silent skip, so adding a new agent tells me about every gap at once.

Just the skills

The skills in skills/ are plain SKILL.md files, so they need neither this CLI nor my settings. Any agent that reads skills can install them straight from this repo:

npx skills add pratikpakhale/pakhale -g                    # all of them
npx skills add pratikpakhale/pakhale -g -s deslop stackup  # or pick

npx pakhale setup agents stays the full path — the same skills plus plugins, MCP servers, instructions and settings.

Options

• plan or -n, --dry-run — show every change, write nothing
• -a, --agent <id> — set up one agent (repeatable)
• --all — every configured agent, no prompt
• --force — resolve every conflict in the config's favour, no prompts
• --home <dir> — apply against a different home directory
• -h, --help — full usage

With a terminal and no --agent, it asks which agents to set up — space to toggle, a for all. Piped or scripted, it applies to every configured agent without asking.

Safe by default

• Re-running changes nothing. Every artifact is checked before it is written, installers included.
• Nothing of yours is overwritten silently. Setup remembers what it last wrote; anything else it finds — an existing CLAUDE.md, a setting changed by hand — is a conflict, and it asks: keep yours, use the config's (backed up first), or see the diff, with one answer for all remaining if you prefer. Piped runs keep yours; --force takes the config's.
• Settings you added yourself survive. Keys config.ts does not name are preserved.
• A config file it cannot parse is left alone rather than flattened.
• Agents you do not have installed are skipped, never created.
• Anything it displaces is backed up to .setup-backups/<timestamp>/.

Development

bun install
bun src/cli.ts setup agents plan   # run from source
bun run test                       # ~5s, no network, never touches your real ~
bun run typecheck
bun run build                      # dist/cli.js (Node >=20)

ARCHITECTURE.md covers how it works inside — the config vocabulary, the plan/diff/commit split, and how to add another agent.