refactor(agent-sync): improve comments and update file handling for agent prompts

chore(package): bump version to 0.1.3

bin/agent-sync.mjs

@@ -1,13 +1,15 @@
 #!/usr/bin/env node
 // agent-sync — installs the bundled skills and agent prompts into the canonical
-// user-level directory (~/.agents) and links them into every detected AI agent's
-// own directory (same architecture as the vercel-labs/skills CLI).
+// 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, existsSync, lstatSync, mkdirSync, readdirSync,
+  cpSync, copyFileSync, existsSync, lstatSync, mkdirSync, readdirSync,
   readFileSync, realpathSync, rmSync, statSync, symlinkSync,
 } from 'node:fs';
 import { homedir, platform } from 'node:os';
@@ -208,10 +210,10 @@ function pointsTo(linkPath, targetPath) {
   }
 }
 
-// Ensures the canonical agents directory exists as a real directory, so the
-// individual .md links have a place to live. 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).
+// 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) {
@@ -220,7 +222,6 @@ function ensureCanonicalAgentsDir() {
     return { label: 'agents', status: 'added' };
   }
   if (current.isSymbolicLink()) {
-    if (!pointsTo(CANONICAL_AGENTS_DIR, SOURCE_AGENTS_DIR)) return { label: 'agents', status: 'skipped' };
     if (DRY_RUN) return { label: 'agents', status: 'relink' };
     rmSync(CANONICAL_AGENTS_DIR, { recursive: true, force: true });
     mkdirSync(CANONICAL_AGENTS_DIR, { recursive: true });
@@ -229,8 +230,9 @@ function ensureCanonicalAgentsDir() {
   return current.isDirectory() ? { label: 'agents', status: 'up-to-date' } : { label: 'agents', status: 'skipped' };
 }
 
-// Copies the selected skills into the canonical directory and links the agent
-// prompt files individually (no copies, ever). Returns the list of actions.
+// 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) {
@@ -244,7 +246,13 @@ function syncCanonical(skills, agentFiles) {
     actions.push({ label: `skills/${name}`, status });
   }
   for (const file of agentFiles) {
-    actions.push({ label: `agents/${file}`, status: ensureLink(join(CANONICAL_AGENTS_DIR, file), join(SOURCE_AGENTS_DIR, file)) });
+    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;
 }

package.json

@@ -1,6 +1,6 @@
 {
   "name": "agent-sync",
-  "version": "0.1.2",
+  "version": "0.1.3",
   "description": "Installs bundled skills and agent prompts into the canonical ~/.agents directory and links them into every detected AI agent (Claude Code, pi, Codex, OpenCode, Cursor, Gemini, Trae, Windsurf, Aider Desk).",
   "type": "module",
   "bin": {