oh-my-skills/bin/agent-sync.mjs

#!/usr/bin/env node
// agent-sync — installs the bundled skills and agent prompts into the canonical
// user-level directory (~/.agents) and links the skills into every detected AI
// agent's own directory (same architecture as the vercel-labs/skills CLI).
// Agent prompts are managed only inside ~/.agents/agents via plain copies
// (file links on windows require privileges).
// Selection is explicit (--all, --skills, --agents, --targets): without a
// selection flag the tool prints its help. See HELP for the full surface.

import { createHash } from 'node:crypto';
import {
  cpSync, copyFileSync, existsSync, lstatSync, mkdirSync, readdirSync,
  readFileSync, realpathSync, rmSync, statSync, symlinkSync,
} from 'node:fs';
import { homedir, platform } from 'node:os';
import { basename, dirname, join, relative } from 'node:path';
import { fileURLToPath } from 'node:url';

const EXIT_OK = 0;
const EXIT_FAILURE = 1;
const EXIT_USAGE = 2;

const IS_WINDOWS = platform() === 'win32';

const PACKAGE_ROOT = join(dirname(fileURLToPath(import.meta.url)), '..');
const SOURCE_SKILLS_DIR = join(PACKAGE_ROOT, 'skills');
const SOURCE_AGENTS_DIR = join(PACKAGE_ROOT, 'agents');

const CANONICAL_SKILLS_DIR = join(homedir(), '.agents', 'skills');
const CANONICAL_AGENTS_DIR = join(homedir(), '.agents', 'agents');

const HELP = `agent-sync — installs bundled skills and agent prompts into the
canonical ~/.agents directory and links them into the AI agents detected
on the machine.

Usage:
  agent-sync --all [--check] [--targets [<names>]]
  agent-sync --skills [<names>] [--agents [<names>]] [--check] [--targets [<names>]]
  agent-sync --list
  agent-sync --help

A selection flag without a value covers its whole dimension: --skills alone
selects every bundled skill, --agents alone every prompt, --targets alone
every detected agent. With a comma-separated value only the listed names are
used. Flags combine: --skills commit --targets claude,pi syncs just the
commit skill into Claude Code and pi. Without any selection flag agent-sync
prints this help.

  --check   dry-run, reports drift without writing anything (exit 1 on drift)

Exit codes: ${EXIT_OK} ok or help, ${EXIT_FAILURE} drift (dry-run) or failure, ${EXIT_USAGE} usage error.`;

// Resolves the Claude Code config dir, honoring CLAUDE_CONFIG_DIR when set.
function claudeBase() {
  return process.env.CLAUDE_CONFIG_DIR?.trim() || join(homedir(), '.claude');
}

// Resolves a fixed directory under the user home (lazy, re-evaluated per run).
function homeDir(dir) {
  return () => join(homedir(), dir);
}

// Supported AI agents: base() proves installation when the directory exists,
// skillsSub/agentsSub are the subdirectories receiving our links. Agent prompts
// are linked only where the convention is verified (claude); pi gets a minimal
// manifest instead (see writePiManifests) because its manifests carry
// per-machine config such as model and thinking.
const AGENT_REGISTRY = [
  { name: 'claude',     base: claudeBase,                    skillsSub: 'skills', agentsSub: 'agents' },
  { name: 'pi',         base: homeDir(join('.pi', 'agent')), skillsSub: 'skills', agentsSub: 'agents' },
  { name: 'codex',      base: homeDir('.codex'),             skillsSub: 'skills' },
  { name: 'opencode',   base: homeDir('.opencode'),          skillsSub: 'skills' },
  { name: 'cursor',     base: homeDir('.cursor'),            skillsSub: 'skills' },
  { name: 'gemini',     base: homeDir('.gemini'),            skillsSub: 'skills' },
  { name: 'trae',       base: homeDir('.trae'),              skillsSub: 'skills' },
  { name: 'windsurf',   base: homeDir('.windsurf'),          skillsSub: 'skills' },
  { name: 'aider-desk', base: homeDir('.aider-desk'),        skillsSub: 'skills' },
];

// Prints a usage error and exits with the usage exit code.
function fail(message) {
  console.error(`${message}\nRun 'agent-sync --help' for usage.`);
  process.exit(EXIT_USAGE);
}

// Returns the names of the bundled skills (directories under skills/).
function bundledSkillNames() {
  return readdirSync(SOURCE_SKILLS_DIR, { withFileTypes: true })
    .filter((entry) => entry.isDirectory())
    .map((entry) => entry.name);
}

// Returns the file names of the bundled agent prompts (*.md under agents/).
function bundledAgentFiles() {
  return readdirSync(SOURCE_AGENTS_DIR).filter((name) => name.endsWith('.md'));
}

// Dimension flags hold three states: null (flag absent), true (whole
// dimension) or an array of names. A repeated flag accumulates names, and a
// valueless occurrence widens the dimension to everything.
function mergeDimension(current, value) {
  if (current === true || value === null) return true;
  return [...(current ?? []), ...value];
}

const DIMENSION_FLAGS = { '--skills': 'skills', '--agents': 'agents', '--targets': 'targets' };

// Parses the command line into a run plan (syntax only: names are validated
// later against the bundled content). Selection flags are --all, --skills,
// --agents and --targets; every other flag is a modifier.
function parseArgs(argv) {
  const plan = { help: false, list: false, dryRun: false, all: false, skills: null, agents: null, targets: null };
  const optionalList = (index, flag) => {
    const value = argv[index + 1];
    if (!value || value.startsWith('--')) return null;
    const names = value.split(',').map((name) => name.trim()).filter(Boolean);
    if (names.length === 0) fail(`error: ${flag} received an empty list of names`);
    return names;
  };
  for (let i = 0; i < argv.length; i += 1) {
    const arg = argv[i];
    if (arg === '--help' || arg === '-h') plan.help = true;
    else if (arg === '--list') plan.list = true;
    else if (arg === '--check') plan.dryRun = true;
    else if (arg === '--all') plan.all = true;
    else if (DIMENSION_FLAGS[arg]) {
      const key = DIMENSION_FLAGS[arg];
      const value = optionalList(i, arg);
      plan[key] = mergeDimension(plan[key], value);
      if (value !== null) i += 1;
    }
    else fail(`error: unknown argument '${arg}'`);
  }
  return plan;
}

// Validates the requested names against the bundled content and the agent
// registry, then resolves the effective selection. A dimension set to true
// expands to everything; prompt names are given without extension (gv-dev)
// and mapped to their bundled files (gv-dev.md).
function resolveSelection(plan) {
  const skillNames = bundledSkillNames();
  const agentFiles = bundledAgentFiles();
  const registryNames = AGENT_REGISTRY.map((agent) => agent.name);
  const pickSkill = (name) => {
    if (!skillNames.includes(name)) fail(`error: unknown skill '${name}' (available: ${skillNames.join(', ')})`);
    return name;
  };
  const pickAgent = (name) => {
    const file = agentFiles.find((candidate) => basename(candidate, '.md') === name);
    if (!file) fail(`error: unknown agent '${name}' (available: ${agentFiles.map((f) => basename(f, '.md')).join(', ')})`);
    return file;
  };
  const pickTarget = (name) => {
    if (!registryNames.includes(name)) fail(`error: unknown target '${name}' (available: ${registryNames.join(', ')})`);
    return name;
  };
  return {
    skills: plan.all || plan.skills === true ? skillNames : (plan.skills ?? []).map(pickSkill),
    agents: plan.all || plan.agents === true ? agentFiles : (plan.agents ?? []).map(pickAgent),
    targets: plan.targets === true || !plan.targets ? null : plan.targets.map(pickTarget),
  };
}

// Prints the bundled catalog plus the AI agents detected on this machine.
function printCatalog() {
  const detected = AGENT_REGISTRY.filter((agent) => existsSync(agent.base())).map((agent) => agent.name);
  console.log(`skills:  ${bundledSkillNames().join(', ')}`);
  console.log(`agents:  ${bundledAgentFiles().map((file) => basename(file, '.md')).join(', ')}`);
  console.log(`targets: ${AGENT_REGISTRY.map((agent) => `${agent.name}${detected.includes(agent.name) ? '*' : ''}`).join(', ')}  (* = detected)`);
}

// Returns the SHA-256 hex digest of a file, or null when the file does not exist.
function fileHash(filePath) {
  if (!existsSync(filePath)) return null;
  return createHash('sha256').update(readFileSync(filePath)).digest('hex');
}

// Returns a stable digest of a directory: sorted list of "relative path:hash"
// for every file, so any content or structure change alters the result.
function dirHash(dirPath) {
  if (!existsSync(dirPath)) return null;
  const parts = [];
  const walk = (dir, prefix) => {
    for (const entry of readdirSync(dir, { withFileTypes: true })) {
      const full = join(dir, entry.name);
      if (entry.isDirectory()) walk(full, `${prefix}${entry.name}/`);
      else parts.push(`${prefix}${entry.name}:${fileHash(full)}`);
    }
  };
  walk(dirPath, '');
  return createHash('sha256').update(parts.sort().join('\n')).digest('hex');
}

// Returns the lstat of a path, or null when it does not exist (broken links included).
function lstatOrNull(path) {
  try {
    return lstatSync(path);
  } catch {
    return null;
  }
}

// Tells whether the entry at linkPath is a symlink resolving to targetPath.
function pointsTo(linkPath, targetPath) {
  try {
    return realpathSync(linkPath) === realpathSync(targetPath);
  } catch {
    return false;
  }
}

// Ensures the canonical agents directory exists as a real directory receiving
// the copied prompt files. An earlier version linked the whole folder with a
// junction: such a link to our own source is unwound here (removing the
// junction never touches the source files).
function ensureCanonicalAgentsDir() {
  const current = lstatOrNull(CANONICAL_AGENTS_DIR);
  if (!current) {
    if (DRY_RUN) return { label: 'agents', status: 'link' };
    mkdirSync(CANONICAL_AGENTS_DIR, { recursive: true });
    return { label: 'agents', status: 'added' };
  }
  if (current.isSymbolicLink()) {
    if (DRY_RUN) return { label: 'agents', status: 'relink' };
    rmSync(CANONICAL_AGENTS_DIR, { recursive: true, force: true });
    mkdirSync(CANONICAL_AGENTS_DIR, { recursive: true });
    return { label: 'agents', status: 'relinked' };
  }
  return current.isDirectory() ? { label: 'agents', status: 'up-to-date' } : { label: 'agents', status: 'skipped' };
}

// Copies the selected skills into the canonical directory and copies the
// agent prompts into ~/.agents/agents (files on windows cannot be linked
// without privileges, so agents use plain copies). Returns the list of actions.
function syncCanonical(skills, agentFiles) {
  const actions = [];
  for (const name of skills) {
    const source = join(SOURCE_SKILLS_DIR, name);
    const dest = join(CANONICAL_SKILLS_DIR, name);
    const status = dirHash(source) === dirHash(dest) ? 'up-to-date' : existsSync(dest) ? 'updated' : 'added';
    if (status !== 'up-to-date' && !DRY_RUN) {
      rmSync(dest, { recursive: true, force: true });
      cpSync(source, dest, { recursive: true });
    }
    actions.push({ label: `skills/${name}`, status });
  }
  for (const file of agentFiles) {
    const source = join(SOURCE_AGENTS_DIR, file);
    const dest = join(CANONICAL_AGENTS_DIR, file);
    const status = fileHash(source) === fileHash(dest) ? 'up-to-date' : existsSync(dest) ? 'updated' : 'added';
    if (status !== 'up-to-date' && !DRY_RUN) {
      copyFileSync(source, dest);
    }
    actions.push({ label: `agents/${file}`, status });
  }
  return actions;
}

// Detects installed agents by probing their config directory, optionally
// restricted to the requested target names.
function detectAgents(targetNames) {
  return AGENT_REGISTRY
    .filter((agent) => (!targetNames || targetNames.includes(agent.name)) && existsSync(agent.base()))
    .map((agent) => ({ ...agent, baseDir: agent.base() }));
}

// Ensures linkPath points to targetPath via a symlink: junctions for
// directories on windows (no privileges needed), file symlinks everywhere
// else. There is no fallback: if the symlink cannot be created (EPERM without
// Developer Mode/privileges, cross-filesystem restrictions) the sync aborts
// with an explicit error — copies are never made. A pre-existing regular file
// is a leftover from older copy/hardlink installs and is replaced by the
// symlink; directories or foreign symlinks are skipped, never touched.
function ensureLink(linkPath, targetPath) {
  const current = lstatOrNull(linkPath);
  if (current) {
    if (current.isSymbolicLink() && pointsTo(linkPath, targetPath)) return 'up-to-date';
    if (current.isFile()) {
      if (DRY_RUN) return 'relink';
      rmSync(linkPath, { force: true });
    } else {
      return 'skipped';
    }
  }
  if (DRY_RUN) return 'link';
  mkdirSync(dirname(linkPath), { recursive: true });
  const isDirectory = statSync(targetPath).isDirectory();
  try {
    if (IS_WINDOWS && isDirectory) symlinkSync(targetPath, linkPath, 'junction');
    else symlinkSync(relative(dirname(linkPath), targetPath), linkPath);
    return 'linked';
  } catch (error) {
    throw new Error(`cannot create symlink '${linkPath}' (no fallback, no copy: ${error.message})`);
  }
}

// Links the selected skills (source package names only, so third-party
// content in the canonical directory is never propagated) into one detected
// agent's directories. Agent prompts are deliberately not linked here: they
// are managed only inside the canonical ~/.agents/agents directory.
function linkForAgent(agent, skills) {
  const actions = [];
  for (const name of skills) {
    actions.push({
      label: `${agent.name}:skills/${name}`,
      status: ensureLink(join(agent.baseDir, agent.skillsSub, name), join(CANONICAL_SKILLS_DIR, name)),
    });
  }
  return actions;
}

// Orchestrates the sync and prints a per-entry report with a final summary.
function main() {
  if (PLAN.help) {
    console.log(HELP);
    process.exit(EXIT_OK);
  }
  if (!existsSync(SOURCE_SKILLS_DIR) || !existsSync(SOURCE_AGENTS_DIR)) {
    console.error(`error: bundled content not found under ${PACKAGE_ROOT}`);
    process.exit(EXIT_FAILURE);
  }
  const hasContent = PLAN.all || PLAN.skills || PLAN.agents;
  if (PLAN.list && (hasContent || PLAN.targets)) fail('error: --list cannot be combined with selection flags');
  if (PLAN.all && (PLAN.skills || PLAN.agents)) fail('error: --all cannot be combined with --skills or --agents');
  if (PLAN.targets && !hasContent) fail('error: --targets requires --all, --skills or --agents');
  if (PLAN.list) {
    printCatalog();
    process.exit(EXIT_OK);
  }
  if (!hasContent) {
    console.log(HELP);
    process.exit(EXIT_OK);
  }
  const { skills, agents: agentFiles, targets } = resolveSelection(PLAN);

  const detected = detectAgents(targets);
  if (targets && detected.length === 0) {
    fail(`error: none of the requested targets is installed (${targets.join(', ')})`);
  }
  if (!DRY_RUN) {
    mkdirSync(CANONICAL_SKILLS_DIR, { recursive: true });
  }

  const report = [ensureCanonicalAgentsDir(), ...syncCanonical(skills, agentFiles)];
  for (const agent of detected) {
    report.push(...linkForAgent(agent, skills));
  }

  for (const { label, status } of report) {
    console.log(`${status.padEnd(12)} ${label}`);
  }
  console.log(`\nagents detected: ${detected.map((agent) => agent.name).join(', ') || 'none'}`);
  const drift = report.filter((entry) => entry.status !== 'up-to-date' && entry.status !== 'skipped').length;
  if (drift === 0) {
    console.log('Everything up to date.');
  } else {
    console.log(DRY_RUN ? `${drift} entr${drift === 1 ? 'y' : 'ies'} out of sync (dry-run, nothing written).` : `${drift} entr${drift === 1 ? 'y' : 'ies'} synced.`);
  }
  if (DRY_RUN && drift > 0) {
    process.exit(EXIT_FAILURE);
  }
}

const PLAN = parseArgs(process.argv.slice(2));
const DRY_RUN = PLAN.dryRun;

try {
  main();
} catch (error) {
  console.error(`error: ${error.message}`);
  process.exit(EXIT_FAILURE);
}