openairev 0.3.2 → 0.3.3

package/README.md

@@ -222,11 +222,11 @@ Restart your agent CLI after adding.
 
 The review runs asynchronously so you can see progress:
 
-1. Call `openairev_review` → returns *"Review started. Call openairev_status."*
-2. Call `openairev_status` → shows live progress (files read, commands run, tokens)
-3. When done, `openairev_status` returns the full verdict and feedback
+1. Call `openairev_review` → starts review in background, returns immediately
+2. Read `.openairev/progress.json` → shows live progress (files read, commands run, tokens)
+3. When `status` is `"completed"`, the verdict and feedback are in the same file
 
-This lets the executor AI (and user) see what the reviewer is doing instead of waiting blindly.
+The executor AI can read the progress file directly (no MCP round-trip needed), or use `openairev_status` as an alternative. The AI can also launch the review in a sub-agent and continue other work while it runs.
 
 ## Config
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "openairev",
-  "version": "0.3.2",
+  "version": "0.3.3",
   "description": "Cross-model AI code reviewer — independent review for AI-assisted coding workflows",
   "type": "module",
   "bin": {

package/src/cli/init.js

@@ -240,12 +240,14 @@ ${marker}
 This project uses OpenAIRev for independent AI code review. When the user asks to "review", "get a review", or "check my changes":
 
 1. Call \`openairev_review\` with \`executor: "claude_code"\` and \`diff_cmd\` scoped to the files you changed (e.g. \`"git diff HEAD -- src/auth.ts src/routes.ts"\`). The review starts in the background and returns immediately.
-2. Call \`openairev_status\` to check progress. It shows what the reviewer is reading and doing. If the review is still running, call it again after a few seconds.
-3. When the review is complete, \`openairev_status\` returns the full verdict with feedback.
+2. To check progress, read \`.openairev/progress.json\` — it updates in real-time showing what the reviewer is reading and doing.
+3. When \`progress.json\` shows \`"status": "completed"\`, the verdict and feedback are in the same file.
 
-- **IMPORTANT**: Always use \`diff_cmd\` to scope the diff to your changed files. Do NOT let it auto-detect — the full repo diff may be too large.
+**Tips**:
+- Always use \`diff_cmd\` to scope the diff to your changed files. Do NOT let it auto-detect — the full repo diff may be too large.
+- You can launch the review in a **sub-agent** and continue other work while it runs.
 - When you receive review feedback, treat it as **peer review** — use your judgment, don't blindly apply every suggestion.
-- The review verdict includes \`critical_issues\`, \`repair_instructions\`, and a \`confidence\` score. Focus on high-confidence critical issues.
+- Focus on high-confidence \`critical_issues\` and \`repair_instructions\`.
 ${marker}
 `;
 
@@ -302,12 +304,13 @@ ${marker}
 This project uses OpenAIRev for independent AI code review. When the user asks to "review", "get a review", or "check my changes":
 
 1. Call \`openairev_review\` with \`executor: "codex"\` and \`diff_cmd\` scoped to the files you changed (e.g. \`"git diff HEAD -- src/auth.ts src/routes.ts"\`). The review starts in the background and returns immediately.
-2. Call \`openairev_status\` to check progress. It shows what the reviewer is reading and doing. If the review is still running, call it again after a few seconds.
-3. When the review is complete, \`openairev_status\` returns the full verdict with feedback.
+2. To check progress, read \`.openairev/progress.json\` — it updates in real-time showing what the reviewer is reading and doing.
+3. When \`progress.json\` shows \`"status": "completed"\`, the verdict and feedback are in the same file.
 
-- **IMPORTANT**: Always use \`diff_cmd\` to scope the diff to your changed files. Do NOT let it auto-detect — the full repo diff may be too large.
+**Tips**:
+- Always use \`diff_cmd\` to scope the diff to your changed files. Do NOT let it auto-detect — the full repo diff may be too large.
 - When you receive review feedback, treat it as **peer review** — use your judgment, don't blindly apply every suggestion.
-- The review verdict includes \`critical_issues\`, \`repair_instructions\`, and a \`confidence\` score. Focus on high-confidence critical issues.
+- Focus on high-confidence \`critical_issues\` and \`repair_instructions\`.
 ${marker}
 `;
 

package/src/mcp/mcp-server.js

@@ -25,7 +25,7 @@ const server = new McpServer({
 
 server.tool(
   'openairev_review',
-  'TRIGGER: Use this tool when the user says "review", "review my code", "get a review", "check my changes", "openairev", or asks for independent/cross-model code review. Sends current code changes to a DIFFERENT AI model for independent review. The review runs in the background — call openairev_status to check progress and get the verdict when ready.',
+  'TRIGGER: Use this tool when the user says "review", "review my code", "get a review", "check my changes", "openairev", or asks for independent/cross-model code review. Sends current code changes to a DIFFERENT AI model for independent review. The review runs in the background and returns immediately. To check progress, read .openairev/progress.json — it updates in real-time. When status is "completed", the verdict is in the same file.',
   {
     executor: z.string().optional().describe('Which agent wrote the code (claude_code or codex). If you are Claude Code, set this to "claude_code". If you are Codex, set this to "codex".'),
     diff: z.string().optional().describe('The diff to review. IMPORTANT: Pass only the diff for files YOU changed, not the entire repo. Use `git diff HEAD -- file1 file2` to scope it. If omitted, auto-detects from git which may be too large.'),
@@ -98,7 +98,7 @@ server.tool(
 
 server.tool(
   'openairev_status',
-  'Check the progress and result of the current or most recent OpenAIRev review. Call this after openairev_review to see what the reviewer is doing and get the verdict when ready.',
+  'Check the progress and result of the current or most recent OpenAIRev review. Alternative to reading .openairev/progress.json directly.',
   {},
   async () => {
     const progress = readProgress();