greenrun-cli 0.1.3 → 0.1.5

package/dist/cli.js

@@ -15,6 +15,7 @@ function printHelp() {
 
 Usage:
   greenrun init             Interactive setup wizard
+  greenrun update           Update command templates to latest version
   greenrun serve            Start MCP server
   greenrun --version, -v    Print version
   greenrun --help, -h       Print this help
@@ -42,6 +43,11 @@ async function main() {
         await runInit(args.slice(1));
         return;
     }
+    if (command === 'update') {
+        const { runUpdate } = await import('./commands/init.js');
+        runUpdate();
+        return;
+    }
     if (command === 'serve') {
         const { startServer } = await import('./server.js');
         await startServer();

package/dist/commands/init.d.ts

@@ -1 +1,2 @@
+export declare function runUpdate(): void;
 export declare function runInit(args: string[]): Promise<void>;

package/dist/commands/init.js

@@ -138,6 +138,56 @@ function installClaudeMd() {
         console.log('  Created CLAUDE.md with Greenrun instructions');
     }
 }
+function installSettings() {
+    const settingsDir = join(process.cwd(), '.claude');
+    mkdirSync(settingsDir, { recursive: true });
+    const settingsPath = join(settingsDir, 'settings.local.json');
+    let existing = {};
+    if (existsSync(settingsPath)) {
+        try {
+            existing = JSON.parse(readFileSync(settingsPath, 'utf-8'));
+        }
+        catch {
+            // overwrite invalid JSON
+        }
+    }
+    const greenrunTools = [
+        'mcp__greenrun__list_projects',
+        'mcp__greenrun__get_project',
+        'mcp__greenrun__create_project',
+        'mcp__greenrun__list_pages',
+        'mcp__greenrun__create_page',
+        'mcp__greenrun__list_tests',
+        'mcp__greenrun__get_test',
+        'mcp__greenrun__create_test',
+        'mcp__greenrun__update_test',
+        'mcp__greenrun__start_run',
+        'mcp__greenrun__complete_run',
+        'mcp__greenrun__get_run',
+        'mcp__greenrun__list_runs',
+        'mcp__greenrun__sweep',
+    ];
+    const browserTools = [
+        'mcp__claude-in-chrome__tabs_context_mcp',
+        'mcp__claude-in-chrome__tabs_create_mcp',
+        'mcp__claude-in-chrome__navigate',
+        'mcp__claude-in-chrome__computer',
+        'mcp__claude-in-chrome__read_page',
+        'mcp__claude-in-chrome__find',
+        'mcp__claude-in-chrome__form_input',
+        'mcp__claude-in-chrome__javascript_tool',
+        'mcp__claude-in-chrome__get_page_text',
+        'mcp__claude-in-chrome__read_console_messages',
+        'mcp__claude-in-chrome__read_network_requests',
+    ];
+    const requiredTools = [...greenrunTools, ...browserTools];
+    existing.permissions = existing.permissions || {};
+    const currentAllow = existing.permissions.allow || [];
+    const merged = [...new Set([...currentAllow, ...requiredTools])];
+    existing.permissions.allow = merged;
+    writeFileSync(settingsPath, JSON.stringify(existing, null, 2) + '\n');
+    console.log('  Updated .claude/settings.local.json with tool permissions');
+}
 function installCommands() {
     const commandsDir = join(process.cwd(), '.claude', 'commands');
     mkdirSync(commandsDir, { recursive: true });
@@ -153,6 +203,13 @@ function installCommands() {
         console.log(`  Installed /${cmd.replace('.md', '')}`);
     }
 }
+export function runUpdate() {
+    console.log('\nGreenrun - Updating templates\n');
+    installCommands();
+    installSettings();
+    installClaudeMd();
+    console.log('\nDone! Templates updated to latest version.\n');
+}
 export async function runInit(args) {
     const opts = parseFlags(args);
     const interactive = !opts.token;
@@ -241,6 +298,7 @@ export async function runInit(args) {
     if (opts.commands) {
         installCommands();
     }
+    installSettings();
     console.log(`
 Done! Restart Claude Code to connect.
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "greenrun-cli",
-  "version": "0.1.3",
+  "version": "0.1.5",
   "description": "CLI and MCP server for Greenrun - browser test management for Claude Code",
   "type": "module",
   "main": "dist/server.js",

package/templates/commands/greenrun.md

@@ -23,17 +23,25 @@ If no argument is given, run all active tests.
 
 If there are no matching active tests, tell the user and stop.
 
-### 3. Execute tests in parallel
+### 3. Pre-fetch test details
+
+Call `get_test` for ALL matching tests **in parallel** (multiple tool calls in one message). This retrieves the full instructions for each test.
+
+Then call `start_run` for ALL tests **in parallel** to get run IDs.
+
+You now have everything needed to launch agents: test name, full instructions, and run_id for each test.
+
+### 4. Execute tests in parallel
 
 Split the test list into batches of size `concurrency` (from the project settings).
 
-For each batch, launch all tests simultaneously using the **Task tool** with `run_in_background: true`. Each background agent receives a prompt containing everything it needs to execute one test independently:
+For each batch, launch all tests simultaneously using the **Task tool** with `run_in_background: true`. Each background agent receives a prompt with the full instructions and run_id embedded — agents do NOT need to call `get_test` or `start_run`.
 
 ```
 For each test in the current batch, call the Task tool with:
 - subagent_type: "general-purpose"
 - run_in_background: true
-- max_turns: 30
+- max_turns: 50
 - model: "sonnet"
 - prompt: (see below)
 ```
@@ -41,33 +49,46 @@ For each test in the current batch, call the Task tool with:
 The prompt for each background agent should be:
 
 ```
-You are executing a single Greenrun browser test. You have access to browser automation tools and Greenrun MCP tools.
-
-**Test: {test_name}** (ID: {test_id})
-
-Step 1: Call `get_test` with test_id "{test_id}" to get full instructions.
-Step 2: Call `start_run` with test_id "{test_id}" to begin - save the returned `run_id`.
-Step 3: Execute the test instructions using browser automation:
-   - Call `tabs_context_mcp` then create a new browser tab for this test
-   - Follow each instruction step exactly as written
-   - The instructions will tell you where to navigate and what to do
-   - Only take a screenshot when you need to verify a visual assertion — not for every navigation or click
-   - When reading page content, prefer `find` or `read_page` with `filter: "interactive"` over full DOM reads
+You are executing a single Greenrun browser test using browser automation tools. Be efficient — minimize tool calls to complete the test as fast as possible.
+
+**Test: {test_name}**
+**Run ID: {run_id}**
+
+## Test Instructions
+
+{paste the full test instructions from get_test here}
+
+## Execution Steps
+
+1. Call `tabs_context_mcp` then create a new browser tab with `tabs_create_mcp`
+2. Follow each test instruction step exactly as written, using these rules to minimize tool calls:
+
+   **Speed rules (critical):**
+   - NEVER take screenshots. Use `read_page` or `find` for all assertions and to locate elements.
+   - Navigate directly to URLs (e.g. `navigate` to `/tokens`) instead of clicking through nav links
+   - Use `javascript_tool` for quick assertions: `document.querySelector('h1')?.textContent` is faster than `read_page` for checking a heading
+   - Use `read_page` with `filter: "interactive"` to verify multiple things in one call rather than separate `find` calls
+   - Use `form_input` with element refs for filling forms — avoid click-then-type sequences
+   - When clicking elements, use `ref` parameter instead of coordinates to avoid needing screenshots
+   - Combine verification: after a page loads, do ONE `read_page` call and check all assertions from that result
+
+   **Reliability rules:**
    - NEVER trigger JavaScript alerts, confirms, or prompts — they block the browser extension entirely. Before clicking delete buttons or other destructive actions, use `javascript_tool` to override: `window.alert = () => {}; window.confirm = () => true; window.prompt = () => null;`
    - If browser tools stop responding (no result or timeout), assume a dialog is blocking — report the error and stop. Do not keep retrying.
    - If you get stuck or a step fails, record the failure and move on — do not retry more than once
-Step 4: Call `complete_run` with:
-   - run_id: the run ID from step 2
+   - If you are redirected to a login page, try using an existing logged-in tab from `tabs_context_mcp` instead of creating a new one
+
+3. Call `complete_run` with:
+   - run_id: "{run_id}"
    - status: "passed" if all checks succeeded, "failed" if any check failed, "error" if execution was blocked
    - result: a brief summary of what happened (include the failure reason if failed/error)
-Step 5: Close the browser tab you created to clean up.
 
 Return a single line summary: {test_name} | {status} | {result_summary}
 ```
 
 After launching all agents in a batch, wait for them all to complete (use `TaskOutput` to collect results) before launching the next batch.
 
-### 4. Summarize results
+### 5. Summarize results
 
 After all batches complete, collect results from all background agents and present a summary table: