[codex] fix extract-knowhow report template generation

extract-knowhow/commands/extract-knowhow.md

@@ -127,6 +127,8 @@ Read full `.jsonl` files. For sessions > 30,000 chars, split into 25,000-char se
 
 **The goal is to extract tacit knowledge — the hard-won intuition, thinking frameworks, and principles that experts carry in their heads but never write down.** Skills should be useful to ANY researcher in the same subdomain, not just the original author.
 
+**Prefer fewer, stronger skills over many weak ones.** If an item is borderline generic, borderline project-specific, or hard to reuse without the original context, skip it.
+
 When extracting from a specific project, always ask: "Would this help a new PhD student entering this field?" If yes, extract it. If it only makes sense in the context of this particular project, generalize it or skip it.
 
 **Generalize:** "For our LiFePO4 simulation, AMIX=0.05 worked" → "For GGA+U calculations on any transition metal oxide with localized d-electrons, reduce AMIX to 0.05"
@@ -163,6 +165,38 @@ Replace specific references with generic descriptions: "our internal dataset" 
 - Standard textbook knowledge with no novel application
 - Any personally identifiable information
 
+### Reuse Quality Bar
+
+Only keep an item if it passes **all** of these checks:
+
+1. **Transferable:** It applies to a recognizable class of problems in the subdomain, not just one project file or one dataset.
+2. **Actionable:** A researcher could do something differently after reading it.
+3. **Replicable:** The reasoning protocol is concrete enough that another researcher could follow it without hidden project context.
+4. **Non-obvious:** It contains judgment, heuristics, failure diagnosis, or tradeoffs that are not just textbook definitions.
+5. **Scoped correctly:** It is neither too broad ("validate results carefully") nor too narrow ("change line 214 in script X").
+
+Reject items that fail any one of these checks.
+
+### Specificity Calibration
+
+- **Too general:** advice that could apply to almost any research project without change. Example: "Check your data quality before analysis."
+- **Too specific:** advice that depends on one dataset, one repository, one file path, or one unpublished internal convention.
+- **Good:** a reusable pattern with a clear trigger condition, action, and scientific rationale. Example: "When land-surface model validation maps look spatially sparse, first verify that remote CSV endpoints returned actual data rather than HTML error pages, because silent fetch failures often masquerade as missing observations."
+
+When in doubt, rewrite toward the "Good" level or skip the item.
+
+### Replicability Requirements
+
+Each accepted item should make the hidden know-how operational:
+
+- `title`: name the problem class or decision point, not a vague theme
+- `description`: state the trigger condition, recommended action, and why it matters scientifically
+- `reasoning_steps`: include 3-7 concrete steps or checks another researcher could actually follow
+- `tools`: include only tools that materially support the workflow, not every tool mentioned in the session
+- `pitfalls`: describe concrete failure modes, not generic warnings
+
+If you cannot write a concrete reasoning protocol or concrete pitfalls, the item is probably not reusable enough to keep.
+
 ### Output per item
 ```json
 {
@@ -229,7 +263,7 @@ Assemble all results into a single JSON object:
 
 Create directory: `mkdir -p ~/.openscientist`
 
-Read the HTML template from the npm package at `templates/report.html` (installed alongside this command). The template contains all CSS, JS, and the interactive UI. Replace the `__REPORT_DATA__` placeholder in the template with the actual JSON data object from Step 6.2.
+Read the HTML template from the npm package at `templates/report.html` (installed alongside this command). The template contains all CSS, JS, and the interactive UI. Replace only the `__REPORT_DATA__` token in the line `const DATA = __REPORT_DATA__;` with the raw JSON data object from Step 6.2, and preserve every other template line exactly.
 
 If the template file is not found, fall back to writing a minimal HTML page that embeds the JSON data and displays it.
 

extract-knowhow/templates/report.html

@@ -17,6 +17,7 @@
 .stat-label{font-size:.8rem;color:#8b949e}
 .author-bar{display:flex;justify-content:center;gap:1.5rem;margin-bottom:2rem;flex-wrap:wrap}
 .author-field{display:flex;align-items:center;gap:.5rem;color:#8b949e;font-size:.9rem}
+.author-field label{white-space:nowrap}
 .author-field input{background:#161b22;border:1px solid #30363d;color:#e6edf3;padding:6px 12px;border-radius:4px;font-size:.9rem;width:200px}
 .project{background:#161b22;border:1px solid #30363d;border-radius:8px;margin-bottom:1.5rem;overflow:hidden}
 .project-header{padding:1rem 1.5rem;background:#1c2129;border-bottom:1px solid #30363d;display:flex;justify-content:space-between;align-items:center;flex-wrap:wrap;gap:.5rem}
@@ -77,9 +78,9 @@ <h1>🏛️ OpenScientist</h1>
     </div>
   </div>
   <div class="author-bar">
-    <div class="author-field">Name: <input id="author-name" type="text" placeholder="e.g. Albert Einstein"></div>
-    <div class="author-field">Institution: <input id="author-inst" type="text" placeholder="e.g. ETH Zürich Physics"></div>
-    <div class="author-field">Role: <select id="author-role" style="background:#161b22;border:1px solid #30363d;color:#e6edf3;padding:6px 12px;border-radius:4px;font-size:.9rem">
+    <div class="author-field"><label for="author-name">Name:</label> <input id="author-name" type="text" placeholder="e.g. Albert Einstein"></div>
+    <div class="author-field"><label for="author-inst">Institution:</label> <input id="author-inst" type="text" placeholder="e.g. ETH Zürich Physics"></div>
+    <div class="author-field"><label for="author-role">Role:</label> <select id="author-role" style="background:#161b22;border:1px solid #30363d;color:#e6edf3;padding:6px 12px;border-radius:4px;font-size:.9rem">
       <option value="">Select...</option>
       <option value="Undergraduate">Undergraduate Student</option>
       <option value="Master's Student">Master's Student</option>
@@ -107,7 +108,7 @@ <h1>🏛️ OpenScientist</h1>
 </div>
 
 <script>
-// __REPORT_DATA__ is replaced by the prompt with actual extracted JSON
+// Insert the extracted report JSON into the DATA assignment below.
 const DATA = __REPORT_DATA__;
 
 const CAT = {
@@ -120,7 +121,7 @@ <h1>🏛️ OpenScientist</h1>
 
 const projects = DATA.projects || [];
 document.getElementById('author-name').value = DATA.author || '';
-document.getElementById('author-inst').value = DATA.email || '';
+document.getElementById('author-inst').value = DATA.institution || '';
 if(DATA.role) document.getElementById('author-role').value = DATA.role;
 document.getElementById('rdate').textContent = DATA.date || new Date().toISOString().slice(0,10);
 document.getElementById('s-proj').textContent = projects.length;

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 +36,15 @@ 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("Prefer fewer, stronger skills over many weak ones."), "Command includes the stronger skill-selection guidance");
+assert(content.includes("Only keep an item if it passes **all** of these checks:"), "Command includes the reuse quality bar");
+assert(content.includes("If you cannot write a concrete reasoning protocol or concrete pitfalls"), "Command enforces replicability requirements");
+
+console.log("\nTest: report template invariants");
+const template = fs.readFileSync(TEMPLATE, "utf-8");
+assert(template.includes("const DATA = __REPORT_DATA__;"), "Template keeps a single DATA placeholder assignment");
+assert(!template.includes("__REPORT_DATA__ is replaced by"), "Template does not repeat the placeholder in comments");
+assert(template.includes("DATA.institution || ''"), "Template maps institution field from DATA.institution");
 
 console.log("\nTest: postuninstall.js");
 execFileSync(process.execPath, [path.join(SCRIPT_DIR, "postuninstall.js")], { stdio: "pipe" });