majlis 0.3.0 → 0.3.2

package/dist/cli.js

@@ -522,13 +522,29 @@ Before building:
 3. Check docs/classification/ for problem taxonomy
 4. Check docs/experiments/ for prior work
 
-During building:
-- Create branch: exp/NNN-description
-- Create experiment log from template
+## Scope Constraint (CRITICAL)
+
+You get ONE attempt per cycle. Your job is:
+1. Read and diagnose \u2014 understand the problem thoroughly
+2. Form ONE hypothesis about what to fix
+3. Implement ONE focused change (not a multi-step debug session)
+4. Run the benchmark ONCE to see the result
+5. Document what you did, what happened, and what you think the next step should be
+6. STOP
+
+Do NOT iterate. Do NOT try multiple approaches. Do NOT debug your own fix.
+If your change doesn't work, document why and let the cycle continue \u2014
+the adversary, critic, and verifier will help diagnose what went wrong.
+The cycle will come back to you with their insights.
+
+If you find yourself wanting to "try one more thing," that's the signal to stop
+and write up what you learned. The other agents exist precisely for this reason.
+
+## During building:
 - Tag EVERY decision: proof / test / strong-consensus / consensus / analogy / judgment
 - When making judgment-level decisions, state: "This is judgment \u2014 reasoning without precedent"
 - Run baseline metrics BEFORE making changes
-- Run comparison metrics AFTER making changes
+- Run comparison metrics AFTER making changes (once)
 
 You may NOT verify your own work or mark your own decisions as proven.
 Output your decisions in structured format so they can be recorded in the database.
@@ -564,9 +580,8 @@ Rules:
 - Every doubt MUST reference evidence. "This feels wrong" is not a doubt.
 - You may NOT suggest fixes. Identify problems only.
 - Focus on judgment and analogy-level decisions first.
-- You may NOT modify any files. Produce a doubt document only.
-
-Write to docs/doubts/NNN-against-experiment-NNN.md
+- You may NOT modify any files. Produce your doubt document as output only.
+- Do NOT attempt to write files. The framework saves your output automatically.
 
 ## Structured Output Format
 <!-- majlis-json
@@ -592,9 +607,8 @@ For each approach the builder takes, ask:
 - Under what conditions do two things the builder treats as distinct become identical?
 
 Produce constructed counterexamples with reasoning.
-Do NOT suggest fixes. Do NOT modify files.
-
-Write to docs/challenges/NNN-against-experiment-NNN.md
+Do NOT suggest fixes. Do NOT modify files. Do NOT attempt to write files.
+The framework saves your output automatically.
 
 ## Structured Output Format
 <!-- majlis-json
@@ -624,7 +638,7 @@ CONTENT CHECK:
 Grade each component: sound / good / weak / rejected
 Grade each doubt/challenge: confirmed / dismissed (with evidence) / inconclusive
 
-Write to docs/verification/NNN-for-experiment-NNN.md
+Produce your verification report as output. The framework saves it automatically.
 
 ## Structured Output Format
 <!-- majlis-json
@@ -658,7 +672,8 @@ Independently propose:
 Compare your decomposition with the existing classification.
 Flag structural divergences \u2014 these are the most valuable signals.
 
-Write to docs/reframes/NNN.md`,
+Produce your reframe document as output. Do NOT attempt to write files.
+The framework saves your output automatically.`,
       compressor: `---
 name: compressor
 model: opus
@@ -705,9 +720,8 @@ For the given experiment:
 Rules:
 - Present findings neutrally. Report each approach on its own terms.
 - Note where external approaches contradict the current one \u2014 these are the most valuable signals.
-- You may NOT modify code or make decisions. Produce a rihla document only.
-
-Write to docs/rihla/NNN-scout-for-experiment-NNN.md
+- You may NOT modify code or make decisions. Produce your rihla document as output only.
+- Do NOT attempt to write files. The framework saves your output automatically.
 
 ## Structured Output Format
 <!-- majlis-json
@@ -2058,6 +2072,9 @@ function determineNextStep(exp, valid, hasDoubts2, hasChallenges2) {
     throw new Error(`Experiment ${exp.slug} is terminal (${exp.status})`);
   }
   const status2 = exp.status;
+  if (status2 === "classified" /* CLASSIFIED */) {
+    return valid.includes("building" /* BUILDING */) ? "building" /* BUILDING */ : valid[0];
+  }
   if (status2 === "built" /* BUILT */ && !hasDoubts2) {
     return valid.includes("doubted" /* DOUBTED */) ? "doubted" /* DOUBTED */ : valid[0];
   }
@@ -2322,7 +2339,7 @@ async function runQuery(opts) {
       cwd: opts.cwd,
       permissionMode: "bypassPermissions",
       allowDangerouslySkipPermissions: true,
-      maxTurns: 50,
+      maxTurns: 30,
       persistSession: false,
       settingSources: ["project"]
     }
@@ -2362,6 +2379,9 @@ async function runQuery(opts) {
     } else if (message.type === "result") {
       if (message.subtype === "success") {
         costUsd = message.total_cost_usd;
+      } else if (message.subtype === "error_max_turns") {
+        costUsd = "total_cost_usd" in message ? message.total_cost_usd : 0;
+        console.warn(`[majlis] Agent hit max turns (${turnCount}). Returning partial output.`);
       } else {
         const errors = "errors" in message ? message.errors?.join("; ") ?? "Unknown error" : "Unknown error";
         throw new Error(`Agent query failed (${message.subtype}): ${errors}`);
@@ -3095,6 +3115,10 @@ async function executeStep(step, exp, root) {
     case "compressed" /* COMPRESSED */:
       await cycle("compress", []);
       break;
+    case "reframed" /* REFRAMED */:
+      updateExperimentStatus(getDb(root), exp.id, "reframed");
+      info(`Reframe acknowledged for ${exp.slug}. Proceeding to build.`);
+      break;
     default:
       warn(`Don't know how to execute step: ${step}`);
   }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "majlis",
-  "version": "0.3.0",
+  "version": "0.3.2",
   "description": "Multi-agent workflow CLI for structured doubt, independent verification, and compressed knowledge",
   "bin": {
     "majlis": "./dist/cli.js"