majlis 0.7.1 → 0.7.2

package/dist/cli.js

@@ -449,7 +449,7 @@ var init_format = __esm({
 function getExtractionSchema(role) {
   switch (role) {
     case "builder":
-      return '{"decisions": [{"description": "string", "evidence_level": "proof|test|strong_consensus|consensus|analogy|judgment", "justification": "string"}]}';
+      return '{"decisions": [{"description": "string", "evidence_level": "proof|test|strong_consensus|consensus|analogy|judgment", "justification": "string"}], "abandon": {"reason": "string", "structural_constraint": "string"}}';
     case "critic":
       return '{"doubts": [{"claim_doubted": "string", "evidence_level_of_claim": "string", "evidence_for_doubt": "string", "severity": "minor|moderate|critical"}]}';
     case "adversary":
@@ -506,7 +506,7 @@ async function extractStructuredData(role, markdown) {
   const tier1 = extractMajlisJsonBlock(markdown);
   if (tier1) {
     const parsed = tryParseJson(tier1);
-    if (parsed) return parsed;
+    if (parsed) return { data: parsed, tier: 1 };
     console.warn(`[majlis] Malformed JSON in <!-- majlis-json --> block for ${role}. Falling back.`);
   } else {
     console.warn(`[majlis] No <!-- majlis-json --> block found in ${role} output. Falling back.`);
@@ -514,15 +514,18 @@ async function extractStructuredData(role, markdown) {
   const tier2 = extractViaPatterns(role, markdown);
   if (tier2 && hasData(tier2)) {
     console.warn(`[majlis] Used regex fallback for ${role}. Review extracted data.`);
-    return tier2;
+    return { data: tier2, tier: 2 };
   }
   console.warn(`[majlis] Regex fallback insufficient for ${role}. Using Haiku extraction.`);
   const tier3 = await extractViaHaiku(role, markdown);
-  if (tier3) return tier3;
+  if (tier3) {
+    console.warn(`[majlis] Tier 3 (Haiku) extraction used for ${role}. Data provenance degraded.`);
+    return { data: tier3, tier: 3 };
+  }
   console.error(
     `[majlis] FAILED to extract structured data from ${role} output. State machine will continue but data is missing. Manual review required.`
   );
-  return null;
+  return { data: null, tier: null };
 }
 function extractMajlisJsonBlock(markdown) {
   const match = markdown.match(/<!--\s*majlis-json\s*\n?([\s\S]*?)-->/);
@@ -592,6 +595,23 @@ function extractViaPatterns(role, markdown) {
     });
   }
   if (doubts.length > 0) result.doubts = doubts;
+  if (role === "builder") {
+    const abandonPattern = /\[ABANDON\]\s*(.+?)(?:\n|$)[\s\S]*?(?:structural.?constraint|Constraint|CONSTRAINT)\s*[:=]\s*(.+?)(?:\n|$)/im;
+    const abandonMatch = markdown.match(abandonPattern);
+    if (abandonMatch) {
+      result.abandon = {
+        reason: abandonMatch[1].trim(),
+        structural_constraint: abandonMatch[2].trim()
+      };
+    }
+    const invalidMatch = markdown.match(/(?:HYPOTHESIS\s+INVALID|HYPOTHESIS\s+IMPOSSIBLE)\s*[:.\-—]\s*(.+?)(?:\n|$)/im);
+    if (invalidMatch && !result.abandon) {
+      result.abandon = {
+        reason: invalidMatch[1].trim(),
+        structural_constraint: "Extracted via regex \u2014 review original document"
+      };
+    }
+  }
   return result;
 }
 async function extractViaHaiku(role, markdown) {
@@ -631,7 +651,7 @@ ${truncated}`;
   }
 }
 function hasData(output) {
-  return !!(output.decisions && output.decisions.length > 0 || output.grades && output.grades.length > 0 || output.doubts && output.doubts.length > 0 || output.challenges && output.challenges.length > 0 || output.findings && output.findings.length > 0 || output.guidance || output.reframe || output.compression_report || output.gate_decision || output.diagnosis);
+  return !!(output.decisions && output.decisions.length > 0 || output.grades && output.grades.length > 0 || output.doubts && output.doubts.length > 0 || output.challenges && output.challenges.length > 0 || output.findings && output.findings.length > 0 || output.guidance || output.reframe || output.compression_report || output.gate_decision || output.diagnosis || output.abandon);
 }
 function validateForRole(role, output) {
   const required = ROLE_REQUIRED_FIELDS[role];
@@ -846,6 +866,8 @@ function buildPreToolUseGuards(role, cwd) {
     const configFile = path2.resolve(cwd, ".majlis", "config.json");
     const dbFile = path2.resolve(cwd, ".majlis", "majlis.db");
     const settingsFile = path2.resolve(cwd, ".claude", "settings.json");
+    const claudeDir = path2.resolve(cwd, ".claude");
+    const agentsDir = path2.resolve(cwd, ".majlis", "agents");
     const configGuard = async (input) => {
       const toolInput = input.tool_input ?? {};
       const filePath = toolInput.file_path ?? "";
@@ -854,6 +876,9 @@ function buildPreToolUseGuards(role, cwd) {
         if (resolved === configFile || resolved === dbFile || resolved === settingsFile) {
           return { decision: "block", reason: `Builder may not modify framework files: ${filePath}` };
         }
+        if (isInsideDir(resolved, claudeDir) || isInsideDir(resolved, agentsDir)) {
+          return { decision: "block", reason: `Builder may not modify agent definitions or framework settings: ${filePath}` };
+        }
       }
       return {};
     };
@@ -867,6 +892,8 @@ function buildPreToolUseGuards(role, cwd) {
     const configFile = path2.resolve(cwd, ".majlis", "config.json");
     const dbFile = path2.resolve(cwd, ".majlis", "majlis.db");
     const settingsFile = path2.resolve(cwd, ".claude", "settings.json");
+    const claudeDir = path2.resolve(cwd, ".claude");
+    const agentsDir = path2.resolve(cwd, ".majlis", "agents");
     const configGuard = async (input) => {
       const toolInput = input.tool_input ?? {};
       const filePath = toolInput.file_path ?? "";
@@ -875,6 +902,9 @@ function buildPreToolUseGuards(role, cwd) {
         if (resolved === configFile || resolved === dbFile || resolved === settingsFile) {
           return { decision: "block", reason: `Verifier may not modify framework files: ${filePath}` };
         }
+        if (isInsideDir(resolved, claudeDir) || isInsideDir(resolved, agentsDir)) {
+          return { decision: "block", reason: `Verifier may not modify agent definitions or framework settings: ${filePath}` };
+        }
       }
       return {};
     };
@@ -947,14 +977,17 @@ ${taskPrompt}`;
   if (artifactPath) {
     console.log(`[${role}] Artifact written to ${artifactPath}`);
   }
-  const structured = await extractStructuredData(role, markdown);
+  const { data: structured, tier: extractionTier } = await extractStructuredData(role, markdown);
   if (structured) {
     const { valid, missing } = validateForRole(role, structured);
     if (!valid) {
       console.warn(`[${role}] Output missing expected fields: ${missing.join(", ")}`);
     }
   }
-  return { output: markdown, structured, truncated };
+  if (extractionTier === 3) {
+    console.warn(`[${role}] WARNING: Structured output was reconstructed by Haiku (tier 3). Data provenance degraded.`);
+  }
+  return { output: markdown, structured, truncated, extractionTier };
 }
 async function spawnSynthesiser(context, projectRoot, opts) {
   const root = projectRoot ?? findProjectRoot() ?? process.cwd();
@@ -982,7 +1015,7 @@ ${taskPrompt}`;
     role: "synthesiser"
   });
   console.log(`[synthesiser] Complete (cost: $${costUsd.toFixed(4)})`);
-  return { output: markdown, structured: { guidance: markdown }, truncated };
+  return { output: markdown, structured: { guidance: markdown }, truncated, extractionTier: null };
 }
 async function spawnRecovery(role, partialOutput, context, projectRoot) {
   const root = projectRoot ?? findProjectRoot() ?? process.cwd();
@@ -1285,7 +1318,8 @@ var init_config = __esm({
       synthesis: 3e4,
       fragility: 15e3,
       experimentDoc: 15e3,
-      deadEnds: 15e3
+      deadEnds: 15e3,
+      experimentLineage: 15e3
     };
   }
 });
@@ -2495,7 +2529,7 @@ function updateSwarmMember(db, swarmRunId, slug, finalStatus, overallGrade, cost
     WHERE swarm_run_id = ? AND experiment_slug = ?
   `).run(finalStatus, overallGrade, costUsd, error, swarmRunId, slug);
 }
-function exportForCompressor(db, maxLength = 3e4) {
+function exportForCompressor(db, maxLength = 5e4) {
   const experiments = listAllExperiments(db);
   const sections = ["# Structured Data Export (from SQLite)\n"];
   sections.push("## Experiments");
@@ -2561,6 +2595,70 @@ function exportForCompressor(db, maxLength = 3e4) {
   }
   return full;
 }
+function exportExperimentLineage(db, subType, maxLength = 15e3) {
+  const experiments = subType ? db.prepare(`SELECT * FROM experiments WHERE sub_type = ? ORDER BY created_at`).all(subType) : listAllExperiments(db);
+  if (experiments.length === 0) return "";
+  const sections = ["## Experiment Lineage (from DB \u2014 canonical, not from synthesis)\n"];
+  for (const exp of experiments) {
+    sections.push(`### ${exp.slug} [${exp.status}]`);
+    if (exp.hypothesis) sections.push(`Hypothesis: ${exp.hypothesis}`);
+    const decisions = listDecisionsByExperiment(db, exp.id);
+    if (decisions.length > 0) {
+      sections.push("Decisions:");
+      for (const d of decisions) {
+        sections.push(`  - [${d.evidence_level}/${d.status}] ${d.description}`);
+      }
+    }
+    const beforeMetrics = getMetricsByExperimentAndPhase(db, exp.id, "before");
+    const afterMetrics = getMetricsByExperimentAndPhase(db, exp.id, "after");
+    if (beforeMetrics.length > 0 && afterMetrics.length > 0) {
+      sections.push("Metrics:");
+      for (const bm of beforeMetrics) {
+        const am = afterMetrics.find((a) => a.fixture === bm.fixture && a.metric_name === bm.metric_name);
+        if (am) {
+          const delta = am.metric_value - bm.metric_value;
+          const sign = delta >= 0 ? "+" : "";
+          sections.push(`  - ${bm.fixture}/${bm.metric_name}: ${bm.metric_value} \u2192 ${am.metric_value} (${sign}${delta.toFixed(4)})`);
+        }
+      }
+    }
+    const doubts = getDoubtsByExperiment(db, exp.id);
+    const resolved = doubts.filter((d) => d.resolution);
+    if (resolved.length > 0) {
+      sections.push("Doubt resolutions:");
+      for (const d of resolved) {
+        sections.push(`  - [${d.resolution}] ${d.claim_doubted}`);
+      }
+    }
+    const verifications = getVerificationsByExperiment(db, exp.id);
+    if (verifications.length > 0) {
+      sections.push("Grades:");
+      for (const v of verifications) {
+        sections.push(`  - ${v.component}: ${v.grade}${v.notes ? ` \u2014 ${v.notes}` : ""}`);
+      }
+    }
+    sections.push("");
+    const current = sections.join("\n");
+    if (current.length > maxLength - 500) {
+      sections.push(`[LINEAGE TRUNCATED \u2014 ${experiments.length - experiments.indexOf(exp) - 1} experiments omitted]`);
+      break;
+    }
+  }
+  const deadEnds = subType ? listDeadEndsBySubType(db, subType) : listAllDeadEnds(db);
+  if (deadEnds.length > 0) {
+    sections.push("### Dead Ends (structural constraints)");
+    for (const de of deadEnds) {
+      sections.push(`- [${de.category ?? "structural"}] ${de.approach}: ${de.structural_constraint}`);
+    }
+  }
+  const full = sections.join("\n");
+  if (full.length > maxLength) {
+    return full.slice(0, maxLength) + `
+
+[LINEAGE TRUNCATED at ${maxLength} chars]`;
+  }
+  return full;
+}
 function exportForDiagnostician(db, maxLength = 6e4) {
   const base = exportForCompressor(db, maxLength);
   const sections = [base];
@@ -3850,9 +3948,18 @@ Output your gate_decision as "approve", "reject", or "flag" with reasoning.`
   const decision = result.structured?.gate_decision ?? "approve";
   const reason = result.structured?.reason ?? "";
   if (decision === "reject") {
-    updateExperimentStatus(db, exp.id, "gated");
-    warn(`Gate REJECTED for ${exp.slug}: ${reason}`);
-    warn(`Revise the hypothesis or run \`majlis revert\` to abandon.`);
+    insertDeadEnd(
+      db,
+      exp.id,
+      exp.hypothesis ?? exp.slug,
+      reason,
+      `Gate rejected: ${reason}`,
+      exp.sub_type,
+      "procedural"
+    );
+    adminTransitionAndPersist(db, exp.id, "gated", "dead_end" /* DEAD_END */, "revert");
+    warn(`Gate REJECTED for ${exp.slug}: ${reason}. Dead-ended.`);
+    return;
   } else {
     if (decision === "flag") {
       warn(`Gate flagged concerns for ${exp.slug}: ${reason}`);
@@ -3909,6 +4016,10 @@ Build the experiment: ${exp.hypothesis}` : `Build the experiment: ${exp.hypothes
   }
   taskPrompt += "\n\nNote: The framework captures metrics automatically. Do NOT claim specific numbers unless quoting framework output.";
   const supplementaryContext = loadExperimentContext(exp, root);
+  const lineage = exportExperimentLineage(db, exp.sub_type);
+  if (lineage) {
+    taskPrompt += "\n\n" + lineage;
+  }
   const result = await spawnAgent("builder", {
     experiment: {
       id: exp.id,
@@ -3927,9 +4038,24 @@ Build the experiment: ${exp.hypothesis}` : `Build the experiment: ${exp.hypothes
     synthesis,
     confirmedDoubts,
     supplementaryContext: supplementaryContext || void 0,
+    experimentLineage: lineage || void 0,
     taskPrompt
   }, root);
   ingestStructuredOutput(db, exp.id, result.structured);
+  if (result.structured?.abandon) {
+    insertDeadEnd(
+      db,
+      exp.id,
+      exp.hypothesis ?? exp.slug,
+      result.structured.abandon.reason,
+      result.structured.abandon.structural_constraint,
+      exp.sub_type,
+      "structural"
+    );
+    adminTransitionAndPersist(db, exp.id, "building", "dead_end" /* DEAD_END */, "revert");
+    info(`Builder abandoned ${exp.slug}: ${result.structured.abandon.reason}`);
+    return;
+  }
   if (result.truncated && !result.structured) {
     warn(`Builder was truncated (hit max turns) without producing structured output.`);
     await spawnRecovery("builder", result.output, {
@@ -3937,6 +4063,25 @@ Build the experiment: ${exp.hypothesis}` : `Build the experiment: ${exp.hypothes
     }, root);
     warn(`Experiment stays at 'building'. Run \`majlis build\` to retry or \`majlis revert\` to abandon.`);
   } else {
+    if (config.build?.pre_measure) {
+      try {
+        const [cmd, ...cmdArgs] = config.build.pre_measure.split(/\s+/);
+        (0, import_node_child_process7.execFileSync)(cmd, cmdArgs, {
+          cwd: root,
+          encoding: "utf-8",
+          timeout: 3e4,
+          stdio: ["pipe", "pipe", "pipe"]
+        });
+      } catch (err) {
+        const errMsg = err instanceof Error ? err.message : String(err);
+        const guidance = `Build verification failed after builder completion. Code may be syntactically broken or incomplete.
+Error: ${errMsg.slice(0, 500)}`;
+        storeBuilderGuidance(db, exp.id, guidance);
+        warn(`Build verification failed for ${exp.slug}. Staying at 'building'.`);
+        warn(`Guidance stored for retry. Run \`majlis build\` to retry.`);
+        return;
+      }
+    }
     if (config.metrics?.command) {
       try {
         const output = (0, import_node_child_process7.execSync)(config.metrics.command, {
@@ -3955,6 +4100,15 @@ Build the experiment: ${exp.hypothesis}` : `Build the experiment: ${exp.hypothes
       }
     }
     gitCommitBuild(exp, root);
+    if (result.extractionTier === 3) {
+      warn(`Builder output extracted via Haiku (tier 3). Data provenance degraded.`);
+      const existing = getBuilderGuidance(db, exp.id) ?? "";
+      storeBuilderGuidance(
+        db,
+        exp.id,
+        existing + "\n[PROVENANCE WARNING] Builder structured output was reconstructed by a secondary model (tier 3). Treat reported decisions with additional scrutiny."
+      );
+    }
     updateExperimentStatus(db, exp.id, "built");
     success(`Build complete for ${exp.slug}. Run \`majlis doubt\` or \`majlis challenge\` next.`);
   }
@@ -4150,6 +4304,15 @@ async function doVerify(db, exp, root) {
   }
   updateExperimentStatus(db, exp.id, "verifying");
   const verifierSupplementaryContext = loadExperimentContext(exp, root);
+  const verifierLineage = exportExperimentLineage(db, exp.sub_type);
+  let verifierTaskPrompt = `Verify experiment ${exp.slug}: ${exp.hypothesis}. Check provenance and content. Test the ${doubts.length} doubt(s) and any adversarial challenges.` + metricsSection + doubtReference;
+  if (verifierLineage) {
+    verifierTaskPrompt += "\n\n" + verifierLineage;
+  }
+  const builderGuidanceForVerifier = getBuilderGuidance(db, exp.id);
+  if (builderGuidanceForVerifier?.includes("[PROVENANCE WARNING]")) {
+    verifierTaskPrompt += "\n\nNote: The builder's structured output was reconstructed by a secondary model (tier 3). Treat reported decisions with additional scrutiny.";
+  }
   const result = await spawnAgent("verifier", {
     experiment: {
       id: exp.id,
@@ -4163,7 +4326,8 @@ async function doVerify(db, exp, root) {
     challenges,
     metricComparisons: metricComparisons.length > 0 ? metricComparisons : void 0,
     supplementaryContext: verifierSupplementaryContext || void 0,
-    taskPrompt: `Verify experiment ${exp.slug}: ${exp.hypothesis}. Check provenance and content. Test the ${doubts.length} doubt(s) and any adversarial challenges.` + metricsSection + doubtReference
+    experimentLineage: verifierLineage || void 0,
+    taskPrompt: verifierTaskPrompt
   }, root);
   ingestStructuredOutput(db, exp.id, result.structured);
   if (result.truncated && !result.structured) {
@@ -5432,6 +5596,7 @@ async function swarm(args) {
     if (summary.bestExperiment && isMergeable(summary.bestExperiment.overallGrade)) {
       const best = summary.bestExperiment;
       info(`Best experiment: ${best.worktree.slug} (${best.overallGrade})`);
+      let merged = false;
       try {
         (0, import_node_child_process10.execFileSync)(
           "git",
@@ -5439,9 +5604,72 @@ async function swarm(args) {
           { cwd: root, encoding: "utf-8", stdio: ["pipe", "pipe", "pipe"] }
         );
         success(`Merged ${best.worktree.slug} into main.`);
+        merged = true;
       } catch {
-        warn(`Git merge of ${best.worktree.slug} failed. Merge manually with:`);
-        info(`  git merge ${best.worktree.branch} --no-ff`);
+        warn(`Git merge of ${best.worktree.slug} failed (conflict). Attempting rebase...`);
+        try {
+          (0, import_node_child_process10.execFileSync)("git", ["merge", "--abort"], { cwd: root, encoding: "utf-8", stdio: ["pipe", "pipe", "pipe"] });
+        } catch {
+        }
+        try {
+          (0, import_node_child_process10.execFileSync)(
+            "git",
+            ["rebase", "main", best.worktree.branch],
+            { cwd: root, encoding: "utf-8", stdio: ["pipe", "pipe", "pipe"] }
+          );
+          info(`Rebase of ${best.worktree.slug} onto main succeeded. Re-verifying gates...`);
+          const config = loadConfig(root);
+          let gatesHold = true;
+          if (config.metrics?.command && best.experiment) {
+            try {
+              const output = (0, import_node_child_process10.execSync)(config.metrics.command, {
+                cwd: root,
+                encoding: "utf-8",
+                timeout: 6e4,
+                stdio: ["pipe", "pipe", "pipe"]
+              }).trim();
+              const parsed = parseMetricsOutput(output);
+              for (const m of parsed) {
+                insertMetric(db, best.experiment.id, "after", m.fixture, m.metric_name, m.metric_value);
+              }
+              const comparisons = compareMetrics(db, best.experiment.id, config);
+              const gateViolations = checkGateViolations(comparisons);
+              if (gateViolations.length > 0) {
+                gatesHold = false;
+                warn(`Gate violations after rebase:`);
+                for (const v of gateViolations) {
+                  warn(`  - ${v.fixture}/${v.metric}: ${v.before} \u2192 ${v.after} (delta: ${v.delta})`);
+                }
+              }
+            } catch {
+              warn("Could not re-capture metrics after rebase. Proceeding cautiously.");
+            }
+          }
+          if (gatesHold) {
+            (0, import_node_child_process10.execFileSync)("git", ["checkout", "main"], { cwd: root, encoding: "utf-8", stdio: ["pipe", "pipe", "pipe"] });
+            (0, import_node_child_process10.execFileSync)(
+              "git",
+              ["merge", "--ff-only", best.worktree.branch],
+              { cwd: root, encoding: "utf-8", stdio: ["pipe", "pipe", "pipe"] }
+            );
+            success(`Merged ${best.worktree.slug} into main (via rebase + ff).`);
+            merged = true;
+          } else {
+            warn(`Gate violations after rebase. NOT merging ${best.worktree.slug}.`);
+            info(`Manual resolution needed:`);
+            info(`  git checkout main && git merge ${best.worktree.branch} --no-ff`);
+          }
+        } catch {
+          try {
+            (0, import_node_child_process10.execFileSync)("git", ["rebase", "--abort"], { cwd: root, encoding: "utf-8", stdio: ["pipe", "pipe", "pipe"] });
+          } catch {
+          }
+          warn(`Rebase of ${best.worktree.slug} also failed. Manual merge required:`);
+          info(`  git merge ${best.worktree.branch} --no-ff`);
+        }
+      }
+      if (!merged) {
+        info(`${best.worktree.slug} was NOT merged automatically.`);
       }
     } else {
       info("No experiment achieved sound/good grade. Nothing merged.");
@@ -5594,6 +5822,7 @@ var init_swarm = __esm({
     init_types2();
     init_spawn();
     init_config();
+    init_metrics();
     init_worktree();
     init_runner();
     init_aggregate();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "majlis",
-  "version": "0.7.1",
+  "version": "0.7.2",
   "description": "Multi-agent workflow CLI for structured doubt, independent verification, and compressed knowledge",
   "bin": {
     "majlis": "./dist/cli.js"