[codex] add single-pass peer review to extract-knowhow

extract-knowhow/commands/extract-knowhow.md

@@ -163,6 +163,27 @@ Replace specific references with generic descriptions: "our internal dataset" 
 - Standard textbook knowledge with no novel application
 - Any personally identifiable information
 
+### Single-Pass Peer Review
+
+After generating each skill item, perform **exactly one** peer review using a different AI agent family. Do not skip this when a partner agent is available, and do not let the item bounce through multiple reviewers.
+
+- If the current agent is **Claude Code**, ask **Codex CLI** or **Gemini** to review.
+- If the current agent is **Codex CLI**, ask **Claude Code** or **Gemini** to review.
+- If the current agent is **Gemini**, ask **Codex CLI** or **Claude Code** to review.
+- Do not request a second review after the first peer responds.
+- Do not ask the reviewer to trigger another reviewer.
+
+The peer review should check:
+- scientific correctness
+- transferability beyond the original project
+- replicability of the reasoning protocol
+- de-identification and privacy
+- whether the item is genuinely worth reusing
+
+If the reviewer finds problems, revise the skill once using that feedback, but **do not** send the revised version out for another review pass.
+
+If no partner agent is available in the current environment, record that explicitly instead of pretending review happened.
+
 ### Output per item
 ```json
 {
@@ -173,7 +194,12 @@ Replace specific references with generic descriptions: "our internal dataset" 
   "reasoning_steps": ["Step 1...", "Step 2..."],
   "tools": ["tool — what it does"],
   "pitfalls": ["Mistake and how to avoid"],
-  "confidence": "high | medium | low"
+  "confidence": "high | medium | low",
+  "peer_review": {
+    "reviewer_agent": "claude-code | codex-cli | gemini | unavailable",
+    "review_status": "approved | revise | needs_external_review",
+    "notes": ["Short review findings"]
+  }
 }
 ```
 

extract-knowhow/templates/report.html

@@ -36,6 +36,8 @@
 .skill-title{font-weight:600;flex:1;min-width:0}
 .skill-conf{font-size:.75rem;color:#8b949e;white-space:nowrap}
 .skill-desc{color:#8b949e;font-size:.9rem;margin:.5rem 0 0 2.25rem}
+.skill-review{color:#8b949e;font-size:.82rem;margin:.35rem 0 0 2.25rem}
+.skill-review strong{color:#e6edf3}
 .expand-btn{font-size:.8rem;color:#58a6ff;cursor:pointer;margin-left:2.25rem;margin-top:.25rem;background:none;border:none;padding:0}
 .expand-btn:hover{text-decoration:underline}
 .gh-btn{font-size:.75rem;background:#238636;color:#fff;border:none;padding:3px 10px;border-radius:4px;cursor:pointer;margin-left:.5rem;text-decoration:none;display:inline-block}
@@ -144,6 +146,13 @@ <h1>🏛️ OpenScientist</h1>
   if(!arr||!arr.length) return '<ul><li><span class="editable" contenteditable="true">N/A</span></li></ul>';
   return '<ul>'+arr.map(item=>`<li><span class="editable ${cls||''}" contenteditable="true">${md(item)}</span></li>`).join('')+'</ul>';
 }
+function normReview(pr){
+  return {
+    reviewer_agent: pr && pr.reviewer_agent ? pr.reviewer_agent : 'unavailable',
+    review_status: pr && pr.review_status ? pr.review_status : 'needs_external_review',
+    notes: pr && Array.isArray(pr.notes) && pr.notes.length ? pr.notes : ['Peer review was not recorded in this run.']
+  };
+}
 
 projects.forEach((proj, pi) => {
   let h = `<div class="project"><div class="project-header">
@@ -154,6 +163,7 @@ <h1>🏛️ OpenScientist</h1>
   proj.skills.forEach((sk, si) => {
     const id = `sk-${pi}-${si}`;
     const cn = sk.category.split('-')[0];
+    const pr = normReview(sk.peer_review);
     h += `<div class="skill" id="${id}-w">
       <div class="skill-top">
         <input type="checkbox" class="skill-cb" id="${id}" checked onchange="tog('${id}')">
@@ -162,6 +172,7 @@ <h1>🏛️ OpenScientist</h1>
         <span class="skill-conf">${esc(sk.confidence)}</span>
       </div>
       <div class="skill-desc">${mkEditable(sk.description, 'ed-desc')}</div>
+      <div class="skill-review"><strong>Peer review:</strong> ${mkEditable(pr.review_status, 'ed-pr-status')} by ${mkEditable(pr.reviewer_agent, 'ed-pr-agent')}</div>
       <div class="skill-actions">
         <button class="expand-btn" onclick="exp('${id}')" style="margin:0">▶ Show details</button>
         <a class="gh-btn" onclick="openIssue(${pi},${si}); return false;" href="#">Submit to GitHub</a>
@@ -171,6 +182,7 @@ <h4>Domain Knowledge</h4>${mkEditable(sk.domain_knowledge, 'ed-dk')}
         <h4>Reasoning Protocol</h4>${mkEditableList(sk.reasoning_steps, 'ed-rs')}
         <h4>Tools</h4>${mkEditableList(sk.tools, 'ed-tools')}
         <h4>Common Pitfalls</h4>${mkEditableList(sk.pitfalls, 'ed-pit')}
+        <h4>Peer Review Notes</h4>${mkEditableList(pr.notes, 'ed-pr-note')}
       </div>
     </div>`;
   });
@@ -213,6 +225,9 @@ <h4>Common Pitfalls</h4>${mkEditableList(sk.pitfalls, 'ed-pit')}
   const rs = readEditableList(detail, 'ed-rs');
   const tools = readEditableList(detail, 'ed-tools');
   const pit = readEditableList(detail, 'ed-pit');
+  const prAgent = readEditable(wrap, 'ed-pr-agent') || normReview(sk.peer_review).reviewer_agent;
+  const prStatus = readEditable(wrap, 'ed-pr-status') || normReview(sk.peer_review).review_status;
+  const prNotes = readEditableList(detail, 'ed-pr-note');
 
   const slug = title.toLowerCase().replace(/[^a-z0-9]+/g,'-').replace(/^-|-$/g,'');
   const subLabel = [proj.domain, proj.subdomain].map(s=>s.split('-').map(w=>w.charAt(0).toUpperCase()+w.slice(1)).join(' ')).join(' / ');
@@ -257,6 +272,12 @@ <h4>Common Pitfalls</h4>${mkEditableList(sk.pitfalls, 'ed-pit')}
 
 ${pit.length ? pit.map(p=>'- '+p).join('\n') : '- N/A'}
 
+## Peer Review
+
+- Reviewer agent: ${prAgent}
+- Review status: ${prStatus}
+${prNotes.length ? prNotes.map(n=>'- '+n).join('\n') : '- Peer review notes not recorded.'}
+
 ## References
 
 - Extracted by /extract-knowhow on ${date}
@@ -272,7 +293,7 @@ <h4>Common Pitfalls</h4>${mkEditableList(sk.pitfalls, 'ed-pit')}
     category: sk.category + ' — ' + catLabel,
     skill_content: skillMd,
     generation_method: 'Generated by /extract-knowhow, then reviewed by me',
-    notes: 'Confidence: ' + sk.confidence
+    notes: 'Confidence: ' + sk.confidence + '; Peer review: ' + prStatus + ' by ' + prAgent
   });
 
   return 'https://github.com/OpenScientists/OpenScientist/issues/new?' + params.toString();

extract-knowhow/tests/test-postinstall.js

@@ -9,6 +9,7 @@ const { execFileSync } = require("child_process");
 const COMMANDS_DIR = path.join(os.homedir(), ".claude", "commands");
 const TARGET = path.join(COMMANDS_DIR, "extract-knowhow.md");
 const SCRIPT_DIR = path.join(__dirname, "..", "scripts");
+const TEMPLATE = path.join(__dirname, "..", "templates", "report.html");
 
 let passed = 0;
 let failed = 0;
@@ -35,6 +36,54 @@ assert(fs.existsSync(TARGET), "Command file exists after install");
 const content = fs.readFileSync(TARGET, "utf-8");
 assert(content.includes("extract-knowhow"), "Command file contains expected content");
 assert(content.startsWith("#"), "Command file starts with markdown header");
+assert(content.includes("perform **exactly one** peer review using a different AI agent family"), "Command requires exactly one peer review");
+assert(content.includes("If the current agent is **Codex CLI**, ask **Claude Code** or **Gemini** to review."), "Command defines Codex peer review routing");
+assert(content.includes("\"peer_review\":"), "Command output schema includes peer review metadata");
+
+console.log("\nTest: report template peer review rendering");
+const template = fs.readFileSync(TEMPLATE, "utf-8");
+assert(template.includes("Peer Review"), "Template includes peer review section");
+assert(template.includes("review_status"), "Template includes peer review status handling");
+const sampleHtml = template.replace(
+  "const DATA = __REPORT_DATA__;",
+  "const DATA = " + JSON.stringify({
+    author: "Test User",
+    email: "test@example.com",
+    total_sessions: 1,
+    date: "2026-04-03",
+    projects: [{
+      name: "Test Project",
+      domain: "computer-science",
+      subdomain: "machine-learning",
+      session_count: 1,
+      skills: [{
+        title: "Test skill",
+        category: "06-coding-and-execution",
+        description: "Test description",
+        domain_knowledge: "Test knowledge",
+        reasoning_steps: ["Step 1"],
+        tools: ["tool — role"],
+        pitfalls: ["pitfall"],
+        confidence: "high",
+        peer_review: {
+          reviewer_agent: "claude-code",
+          review_status: "approved",
+          notes: ["Looks reusable."]
+        }
+      }]
+    }]
+  }) + ";"
+);
+const scriptMatch = sampleHtml.match(/<script>([\s\S]*)<\/script>/);
+assert(Boolean(scriptMatch), "Template contains a script block");
+if (scriptMatch) {
+  try {
+    new Function(scriptMatch[1]);
+    assert(true, "Template script parses after peer review data injection");
+  } catch (err) {
+    assert(false, "Template script parses after peer review data injection");
+  }
+}
 
 console.log("\nTest: postuninstall.js");
 execFileSync(process.execPath, [path.join(SCRIPT_DIR, "postuninstall.js")], { stdio: "pipe" });