@trygentic/agentloop 0.13.2-alpha.11 → 0.15.0-alpha.11

package/README.md

@@ -484,7 +484,6 @@ AgentLoop uses YAML configuration files with a layered priority system.
 ```yaml
 orchestrator:
   max_parallel_agents: 1        # Number of agents that can run in parallel
-  use_worktrees: false          # Enable git worktrees for parallel development
   worktrees_dir: ".worktrees"   # Directory for worktrees (relative to project)
   cleanup_worktrees_on_complete: false  # Auto-cleanup worktrees after completion
 ```
@@ -492,7 +491,6 @@ orchestrator:
 | Option | Default | Env Variable | Description |
 |--------|---------|--------------|-------------|
 | `max_parallel_agents` | `1` | `AGENTLOOP_MAX_PARALLEL_AGENTS` | Maximum parallel agents |
-| `use_worktrees` | `false` | `AGENTLOOP_USE_WORKTREES` | Enable git worktrees |
 | `worktrees_dir` | `.worktrees` | `AGENTLOOP_WORKTREES_DIR` | Worktrees directory |
 | `cleanup_worktrees_on_complete` | `false` | `AGENTLOOP_CLEANUP_WORKTREES` | Auto-cleanup worktrees |
 
@@ -576,7 +574,6 @@ github:
 
 orchestrator:
   max_parallel_agents: 4
-  use_worktrees: true
 ```
 
 **Project-level overrides** (`./.agentloop/config.yaml`):
@@ -636,20 +633,6 @@ agentloop config
 - npm or yarn
 - Podman (optional, for container sandboxing)
 
-## Architecture
-
-AgentLoop uses [OpenCode](https://github.com/anthropics/opencode) as a git submodule for:
-
-- **Tool System** (`@opencode/tool/`): Built-in tools (bash, read, write, edit, grep, glob, etc.)
-- **LLM Abstraction** (`@opencode/session/llm`): Vercel AI SDK integration with streaming support
-- **Provider Management** (`@opencode/provider/`): 18+ bundled AI SDK providers
-
-### Core Technologies
-
-- **OCLIF**: CLI framework for command-line interface
-- **OpenTUI**: Terminal UI framework (React-like components)
-- **Better SQLite3**: Local database for state management
-- **MCP (Model Context Protocol)**: Extensible tool system
 
 ### Supported Providers
 
@@ -658,58 +641,6 @@ All providers supported by OpenCode (via Vercel AI SDK):
 - **OpenAI** - Codex models
 - **OpenCode Zen** - Free tier for getting started
 
-
-## MCP Servers
-
-AgentLoop integrates with various Model Context Protocol (MCP) servers to enhance agent capabilities:
-
-### Configured MCP Servers
-
-| Server | Description | Tools |
-|--------|-------------|-------|
-| **serena** | Code editing and codebase traversal | File operations, symbol manipulation, code generation |
-| **code-graph-rag** | Semantic code analysis with RAG | 24 tools for code search, similarity detection, impact analysis |
-| **playwright** | Browser automation and testing | Web scraping, UI testing, interaction automation |
-| **supabase** | Database operations | SQL execution, migrations, schema management |
-| **mcp-atlassian** | Jira integration | Issue tracking, project management |
-| **agentloop-debug** | Internal debugging tools | Task management, agent coordination |
-
-### Code Graph RAG MCP
-
-The **code-graph-rag-mcp** server provides advanced semantic code analysis:
-
-- **Semantic Search**: Find code by meaning, not just keywords
-- **Code Similarity**: Detect duplicate code and similar patterns
-- **Impact Analysis**: Assess the impact of potential changes
-- **Performance**: 5.5x faster than traditional file-based search
-- **Languages**: Supports TypeScript, JavaScript, Python, Java, Go, Rust, C++, C#, Ruby, PHP
-
-**Requirements**: Node.js 18+, 2GB RAM minimum (8GB recommended)
-
-For detailed information, see [docs/mcp/CODE_GRAPH_RAG_MCP.md](./docs/mcp/CODE_GRAPH_RAG_MCP.md)
-
-### Configuring MCP Servers
-
-MCP servers are configured in `.mcp.json`:
-
-```json
-{
-  "mcpServers": {
-    "code-graph-rag": {
-      "command": "npx",
-      "args": ["-y", "@er77/code-graph-rag-mcp", "/work"],
-      "env": {
-        "MCP_SEMANTIC_WARMUP_LIMIT": "1000"
-      }
-    }
-  }
-}
-```
-
-For more MCP integration examples, see [docs/mcp/GITHUB_MCP_INTEGRATION.md](./docs/mcp/GITHUB_MCP_INTEGRATION.md)
-
-
-
 ## Templates
 
 Built-in agent templates are distributed as part of the npm package in the `.claude/agents/` directory. When you install AgentLoop, these templates are automatically available and loaded by the agent system.

package/bin/agentloop

@@ -46,15 +46,23 @@ if [ "$PLATFORM" = "linux" ]; then
 fi
 
 PACKAGE_NAME="@trygentic/agentloop-${PLATFORM}-${ARCH}${SUFFIX}"
-NODE_MODULES="$DIR/../node_modules"
-PLATFORM_BIN="$NODE_MODULES/$PACKAGE_NAME/bin/agentloop"
 
-if [ -x "$PLATFORM_BIN" ]; then
-  exec "$PLATFORM_BIN" "$@"
-fi
+# Try multiple node_modules locations (npm may hoist dependencies)
+# 1. Nested: node_modules/@trygentic/agentloop/node_modules/@trygentic/agentloop-darwin-arm64
+# 2. Hoisted: node_modules/@trygentic/agentloop-darwin-arm64
+NESTED_NODE_MODULES="$DIR/../node_modules"
+HOISTED_NODE_MODULES="$DIR/../../.."
+
+for NODE_MODULES in "$NESTED_NODE_MODULES" "$HOISTED_NODE_MODULES"; do
+  PLATFORM_BIN="$NODE_MODULES/$PACKAGE_NAME/bin/agentloop"
+  if [ -x "$PLATFORM_BIN" ]; then
+    exec "$PLATFORM_BIN" "$@"
+  fi
+done
 
 echo "Error: Could not find agentloop executable" >&2
 echo "Tried:" >&2
 echo "  - $DIR/agentloop-cli (local development)" >&2
-echo "  - $PLATFORM_BIN (npm package)" >&2
+echo "  - $NESTED_NODE_MODULES/$PACKAGE_NAME/bin/agentloop (nested)" >&2
+echo "  - $HOISTED_NODE_MODULES/$PACKAGE_NAME/bin/agentloop (hoisted)" >&2
 exit 1

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@trygentic/agentloop",
-  "version": "0.13.2-alpha.11",
+  "version": "0.15.0-alpha.11",
   "description": "AI-powered autonomous coding agent",
   "bin": {
     "agentloop": "./bin/agentloop"
@@ -9,8 +9,8 @@
     "postinstall": "node ./scripts/postinstall.mjs"
   },
   "optionalDependencies": {
-    "@trygentic/agentloop-darwin-arm64": "0.13.2-alpha.11",
-    "@trygentic/agentloop-linux-x64": "0.13.2-alpha.11"
+    "@trygentic/agentloop-darwin-arm64": "0.15.0-alpha.11",
+    "@trygentic/agentloop-linux-x64": "0.15.0-alpha.11"
   },
   "engines": {
     "node": ">=18.0.0"

package/scripts/postinstall.mjs

@@ -3,8 +3,11 @@
 /**
  * AgentLoop postinstall script
  *
- * Creates symlinks from the main package's bin to the platform executables
- * for faster execution (avoids the Node.js shim overhead).
+ * 1. Creates symlinks from the main package's bin to the platform executables
+ *    for faster execution (avoids the Node.js shim overhead).
+ *
+ * 2. Sets up plugin agents in ~/.agentloop/agents/ with folder structure.
+ *    Plugin agents are non-core agents that can be customized by users.
  *
  * If symlink creation fails (e.g., on Windows), the shim will handle
  * executable discovery at runtime.
@@ -54,6 +57,22 @@ for (const binary of binaries) {
   const sourcePath = path.join(nodeModules, packageName, "bin", executableName);
   const targetPath = path.join(binDir, executableName);
 
+  // Check if local binary already exists (not a symlink)
+  // This happens when installing from a tarball that includes local binaries
+  // In this case, we should NOT replace them with symlinks to npm packages
+  if (fs.existsSync(targetPath)) {
+    try {
+      const stats = fs.lstatSync(targetPath);
+      if (!stats.isSymbolicLink()) {
+        // Target is a real file, not a symlink - preserve it
+        console.log(`agentloop: keeping local ${binary} binary`);
+        continue;
+      }
+    } catch {
+      // If we can't stat, proceed with replacement
+    }
+  }
+
   if (fs.existsSync(sourcePath)) {
     try {
       // Remove existing file/symlink
@@ -80,3 +99,139 @@ if (linked > 0) {
   console.log(`agentloop: platform executables not found (${packageName})`);
   console.log("agentloop: will use runtime executable resolution");
 }
+
+// =============================================================================
+// Global Agents Setup
+// =============================================================================
+// All agents (core and plugin) are copied to ~/.agentloop/agents/
+// in folder-based structure: agents/<name>/<name>.md (and .bt.json if exists)
+// Users can customize any agent by editing the files in their home directory.
+//
+// Core agents: orchestrator, engineer, product-manager, qa-tester
+// Plugin agents: code-review, container,
+//                dag-planner, maestro-qa, merge-resolver, project-detection, questionnaire
+
+const globalAgentsDir = path.join(os.homedir(), ".agentloop", "agents");
+const templatesDir = path.resolve(__dirname, "..", "templates", "agents");
+const behaviorTreesDir = path.resolve(__dirname, "..", "behavior-trees", "agents");
+
+/**
+ * Setup all agents in ~/.agentloop/agents/
+ *
+ * Discovers agents from:
+ * 1. Folder-based structure: templates/agents/<name>/<name>.md (core agents)
+ * 2. Flat files: templates/agents/<name>.md (plugin agents)
+ * 3. Behavior trees: behavior-trees/agents/<name>.bt.json (standalone BTs)
+ */
+function setupGlobalAgents() {
+  // Create global agents directory if it doesn't exist
+  if (!fs.existsSync(globalAgentsDir)) {
+    try {
+      fs.mkdirSync(globalAgentsDir, { recursive: true });
+    } catch (err) {
+      console.log(`agentloop: could not create ${globalAgentsDir}: ${err.message}`);
+      return 0;
+    }
+  }
+
+  // Discover all agents from templates directory
+  const agentsToInstall = new Map(); // agentName -> { mdSource, btSource }
+
+  if (fs.existsSync(templatesDir)) {
+    const entries = fs.readdirSync(templatesDir, { withFileTypes: true });
+
+    for (const entry of entries) {
+      // Skip internal agents directory
+      if (entry.name === "internal") continue;
+
+      if (entry.isDirectory()) {
+        // Folder-based agent: templates/agents/<name>/<name>.md
+        const agentName = entry.name;
+        const mdPath = path.join(templatesDir, agentName, `${agentName}.md`);
+        const btPath = path.join(templatesDir, agentName, `${agentName}.bt.json`);
+
+        agentsToInstall.set(agentName, {
+          mdSource: fs.existsSync(mdPath) ? mdPath : null,
+          btSource: fs.existsSync(btPath) ? btPath : null,
+        });
+      } else if (entry.isFile() && entry.name.endsWith(".md")) {
+        // Flat file agent: templates/agents/<name>.md
+        const agentName = entry.name.replace(".md", "");
+        if (!agentsToInstall.has(agentName)) {
+          agentsToInstall.set(agentName, {
+            mdSource: path.join(templatesDir, entry.name),
+            btSource: null,
+          });
+        }
+      }
+    }
+  }
+
+  // Also check for standalone behavior trees
+  if (fs.existsSync(behaviorTreesDir)) {
+    const btFiles = fs.readdirSync(behaviorTreesDir).filter(f => f.endsWith(".bt.json"));
+    for (const btFile of btFiles) {
+      const agentName = btFile.replace(".bt.json", "");
+      const btPath = path.join(behaviorTreesDir, btFile);
+
+      if (agentsToInstall.has(agentName)) {
+        // Add BT to existing agent if it doesn't already have one
+        const existing = agentsToInstall.get(agentName);
+        if (!existing.btSource) {
+          existing.btSource = btPath;
+        }
+      } else {
+        // Standalone BT (no markdown)
+        agentsToInstall.set(agentName, {
+          mdSource: null,
+          btSource: btPath,
+        });
+      }
+    }
+  }
+
+  let installed = 0;
+
+  for (const [agentName, sources] of agentsToInstall) {
+    const agentDir = path.join(globalAgentsDir, agentName);
+    const mdTarget = path.join(agentDir, `${agentName}.md`);
+    const btTarget = path.join(agentDir, `${agentName}.bt.json`);
+
+    // Skip if agent directory already exists (user may have customized it)
+    if (fs.existsSync(agentDir)) {
+      continue;
+    }
+
+    // Skip if no source files
+    if (!sources.mdSource && !sources.btSource) {
+      continue;
+    }
+
+    try {
+      // Create agent directory
+      fs.mkdirSync(agentDir, { recursive: true });
+
+      // Copy markdown file if exists
+      if (sources.mdSource) {
+        fs.copyFileSync(sources.mdSource, mdTarget);
+      }
+
+      // Copy behavior tree if exists
+      if (sources.btSource) {
+        fs.copyFileSync(sources.btSource, btTarget);
+      }
+
+      installed++;
+    } catch (err) {
+      console.log(`agentloop: failed to install agent ${agentName}: ${err.message}`);
+    }
+  }
+
+  return installed;
+}
+
+// Run global agent setup
+const agentCount = setupGlobalAgents();
+if (agentCount > 0) {
+  console.log(`agentloop: installed ${agentCount} agent(s) to ${globalAgentsDir}`);
+}