@nathapp/nax 0.68.3 → 0.68.5

package/dist/nax.js

@@ -16837,6 +16837,7 @@ var init_schemas_execution = __esm(() => {
     fullSuiteTimeoutSeconds: exports_external.number().int().min(10).max(600).default(120),
     maxFailureSummaryChars: exports_external.number().int().min(500).max(1e4).default(2000),
     abortOnIncreasingFailures: exports_external.boolean().default(true),
+    consecutiveIncreasesToBail: exports_external.number().int().min(1).max(10).default(2),
     escalateOnExhaustion: exports_external.boolean().optional().default(true),
     rethinkAtAttempt: exports_external.number().int().min(1).default(2),
     urgencyAtAttempt: exports_external.number().int().min(1).default(3)
@@ -17321,6 +17322,7 @@ var init_schemas3 = __esm(() => {
         fullSuiteTimeoutSeconds: 300,
         maxFailureSummaryChars: 2000,
         abortOnIncreasingFailures: true,
+        consecutiveIncreasesToBail: 2,
         escalateOnExhaustion: true,
         rethinkAtAttempt: 2,
         urgencyAtAttempt: 3
@@ -19086,6 +19088,12 @@ function resolveTestStrategy(raw) {
     return "three-session-tdd-lite";
   return "test-after";
 }
+function isThreeSessionStrategy(strategy) {
+  return strategy !== undefined && THREE_SESSION_STRATEGIES.has(strategy);
+}
+function isSingleSessionTestOwningStrategy(strategy) {
+  return strategy !== undefined && SINGLE_SESSION_TEST_OWNING_STRATEGIES.has(strategy);
+}
 function getAcQualityRules(profile) {
   const langSection = profile?.language ? LANGUAGE_PATTERNS[profile.language] : undefined;
   const typeSection = profile?.type ? TYPE_PATTERNS[profile.type] : undefined;
@@ -19098,7 +19106,7 @@ function getAcQualityRules(profile) {
 
 ${extras}`;
 }
-var VALID_TEST_STRATEGIES, COMPLEXITY_GUIDE = `## Complexity Classification Guide
+var VALID_TEST_STRATEGIES, THREE_SESSION_STRATEGIES, SINGLE_SESSION_TEST_OWNING_STRATEGIES, COMPLEXITY_GUIDE = `## Complexity Classification Guide
 
 Classify each story's complexity based on scope and risk \u2014 NOT acceptance criteria count.
 A story with 10 simple "add field" ACs is simpler than one with 3 ACs involving concurrent
@@ -19239,6 +19247,11 @@ var init_test_strategy = __esm(() => {
     "three-session-tdd",
     "three-session-tdd-lite"
   ];
+  THREE_SESSION_STRATEGIES = new Set([
+    "three-session-tdd",
+    "three-session-tdd-lite"
+  ]);
+  SINGLE_SESSION_TEST_OWNING_STRATEGIES = new Set(["tdd-simple", "test-after"]);
   LANGUAGE_PATTERNS = {
     go: `### Go-Specific AC Patterns
 
@@ -36046,7 +36059,7 @@ var init_autofix_test_writer = __esm(() => {
     kind: "run",
     name: "autofix-test-writer",
     stage: "rectification",
-    session: { role: "test-writer", lifetime: "fresh" },
+    session: { role: "test-writer", lifetime: "warm" },
     config: autofixConfigSelector,
     build(input, _ctx) {
       const prompt = RectifierPromptBuilder.testWriterRectification(input.failedChecks, input.story, {
@@ -36838,18 +36851,36 @@ var init__session_output = __esm(() => {
   EMPTY = { success: false, filesChanged: [], output: "", parsed: false };
 });
 
+// src/operations/execution-gates.ts
+function shouldRunReview(config2) {
+  return config2.review?.enabled === true;
+}
+function shouldRunRectification(config2) {
+  return config2.execution?.rectification?.enabled === true;
+}
+function shouldKeepSessionOpen(config2, role) {
+  return SESSION_CONTINUITY_ROLES.has(role) && (shouldRunReview(config2) || shouldRunRectification(config2));
+}
+var SESSION_CONTINUITY_ROLES;
+var init_execution_gates = __esm(() => {
+  init_config();
+  SESSION_CONTINUITY_ROLES = new Set(["implementer", "test-writer"]);
+});
+
 // src/operations/write-test.ts
 var testWriterOp;
 var init_write_test = __esm(() => {
   init_config();
   init_isolation();
   init__session_output();
+  init_execution_gates();
   testWriterOp = {
     kind: "run",
     name: "test-writer",
     stage: "run",
-    session: { role: "test-writer", lifetime: "fresh" },
+    session: { role: "test-writer", lifetime: "warm" },
     config: tddConfigSelector,
+    keepOpen: (_input, ctx) => shouldKeepSessionOpen(ctx.config, "test-writer"),
     build(input, _ctx) {
       if (input.promptMarkdown?.trim()) {
         return {
@@ -36896,20 +36927,6 @@ var init_write_test = __esm(() => {
   };
 });
 
-// src/operations/execution-gates.ts
-function shouldRunReview(config2) {
-  return config2.review?.enabled === true;
-}
-function shouldRunRectification(config2) {
-  return config2.execution?.rectification?.enabled === true;
-}
-function shouldKeepSessionOpen(config2, role) {
-  return role === "implementer" && (shouldRunReview(config2) || shouldRunRectification(config2));
-}
-var init_execution_gates = __esm(() => {
-  init_config();
-});
-
 // src/operations/implement.ts
 var implementerOp;
 var init_implement = __esm(() => {
@@ -38416,10 +38433,11 @@ var init__finding_to_check = __esm(() => {
 });
 
 // src/operations/autofix-implementer-strategy.ts
-function makeAutofixImplementerStrategy(story, config2, sink) {
+function makeAutofixImplementerStrategy(story, config2, sink, opts = {}) {
+  const claimsAdversarial = opts.includeAdversarialReview === true;
   return {
     name: "autofix-implementer",
-    appliesTo: (f) => f.fixTarget === "source" && IMPLEMENTER_SOURCES.has(f.source),
+    appliesTo: (f) => f.fixTarget === "source" && IMPLEMENTER_SOURCES.has(f.source) || claimsAdversarial && f.source === "adversarial-review",
     fixOp: implementerRectifyOp,
     buildInput: (findings, _prior, _cycleCtx) => ({
       failedChecks: findingsToFailedChecks(findings),
@@ -41077,11 +41095,19 @@ ${exceptions.join(`
 
 `)}`;
 }
+function implementerOwnsTests(story) {
+  return isSingleSessionTestOwningStrategy(story.routing?.testStrategy);
+}
+function testEditHeadline(story, prohibition) {
+  return implementerOwnsTests(story) ? SINGLE_SESSION_PERMIT_HEADLINE : prohibition;
+}
 function exceptionCountWord(story) {
-  return THREE_SESSION_STRATEGIES.has(story.routing?.testStrategy ?? "") ? "four" : "three";
+  return isThreeSessionStrategy(story.routing?.testStrategy) ? "four" : "three";
 }
 function escapeHatchFor(story) {
-  const isTdd = THREE_SESSION_STRATEGIES.has(story.routing?.testStrategy ?? "");
+  if (implementerOwnsTests(story))
+    return SINGLE_SESSION_TEST_EDIT_POLICY;
+  const isTdd = isThreeSessionStrategy(story.routing?.testStrategy);
   return buildEscapeHatch({ includeMockHandoff: isTdd });
 }
 function noTestIsolationBlock(story) {
@@ -41152,7 +41178,7 @@ ${errors3}
 2. Only fix findings that are actually valid problems
 3. Do NOT add keys, functions, or imports that already exist \u2014 check first
 
-Do NOT change test files or test behavior \u2014 see the ${exceptionCountWord(story)} narrow exceptions appended below.
+${testEditHeadline(story, `Do NOT change test files or test behavior \u2014 see the ${exceptionCountWord(story)} narrow exceptions appended below.`)}
 Do NOT add new features \u2014 only fix valid issues.
 Commit your fixes when done.${scopeConstraint}${noTestIsolationBlock(story)}${escapeHatchFor(story)}`;
 }
@@ -41206,6 +41232,7 @@ Commit your fixes when done.${scopeConstraint}${noTestIsolationBlock(story)}${es
 }
 function mechanicalRectification(checks3, story, scopeConstraint, opts) {
   const errors3 = formatCheckErrors(checks3, opts);
+  const scopeDirective = implementerOwnsTests(story) ? `Fix all errors listed above that are within this story's scope. ${SINGLE_SESSION_PERMIT_HEADLINE}` : `Fix all errors listed above that are within this story's scope \u2014 see the ${exceptionCountWord(story)} narrow exceptions appended below for sibling-story spillover. Do NOT change test files or test behavior except via those exceptions.`;
   return `You are fixing lint/typecheck errors from a code review.
 
 Story: ${story.title} (${story.id})
@@ -41214,7 +41241,7 @@ The following quality checks failed after implementation:
 
 ${errors3}
 
-Fix all errors listed above that are within this story's scope \u2014 see the ${exceptionCountWord(story)} narrow exceptions appended below for sibling-story spillover. Do NOT change test files or test behavior except via those exceptions.
+${scopeDirective}
 Do NOT add new features \u2014 only fix the quality check errors.
 After fixing, re-run the failing check(s) to verify they pass, then commit your changes.${scopeConstraint}${noTestIsolationBlock(story)}${escapeHatchFor(story)}`;
 }
@@ -41285,11 +41312,28 @@ REASON: <one paragraph: which mock is wrong vs which dispatch the new code uses,
 Rules:
 - Do NOT make any edits yourself; the test-writer will fulfill.
 - Do NOT also emit \`UNRESOLVED:\` in the same turn \u2014 this declaration IS the handoff.
-- FILES must list real test files. Each path must exist and be a test file.`, THREE_SESSION_STRATEGIES, CONTRADICTION_ESCAPE_HATCH, MAX_STRUCTURED_FINDINGS = 10, RAW_WITH_FINDINGS_LIMIT = 1000, RAW_FALLBACK_LIMIT = 4000;
+- FILES must list real test files. Each path must exist and be a test file.`, SINGLE_SESSION_PERMIT_HEADLINE = "You authored these tests in the same session as the implementation, so you MAY edit test files \u2014 but ONLY to resolve a genuine contradiction between a test and this story's acceptance criteria (or between two acceptance criteria). NEVER weaken, delete, loosen, or skip a test merely to make it pass. See the test-edit guidance appended below.", SINGLE_SESSION_TEST_EDIT_POLICY = `
+
+## Test-edit guidance (single-session implementer)
+
+You wrote both the tests and the implementation for this story in one session, so no
+separate test-writer owns the test contract. You therefore MAY edit test files during
+rectification \u2014 subject to these limits:
+
+- Edit a test ONLY to resolve a genuine contradiction between the test and an acceptance
+  criterion, a contradiction between two acceptance criteria, or a clear defect in a test
+  you authored (wrong arity/type, impossible setup, or asserting behavior the ACs do not require).
+- NEVER weaken, delete, loosen, or \`skip\` a test simply because the implementation fails it.
+  A failing test usually means the SOURCE is wrong \u2014 fix the source first.
+- The semantic and adversarial reviewers still gate correctness; gaming a test to pass will be caught.
+
+If two findings or two acceptance criteria contradict each other and you cannot satisfy
+both even after adjusting tests, do not guess. Emit:
+UNRESOLVED: <which findings/ACs conflicted and why they cannot both be satisfied>`, CONTRADICTION_ESCAPE_HATCH, MAX_STRUCTURED_FINDINGS = 10, RAW_WITH_FINDINGS_LIMIT = 1000, RAW_FALLBACK_LIMIT = 4000;
 var init_rectifier_builder_helpers = __esm(() => {
+  init_config();
   init_review();
   init_sections2();
-  THREE_SESSION_STRATEGIES = new Set(["three-session-tdd", "three-session-tdd-lite"]);
   CONTRADICTION_ESCAPE_HATCH = buildEscapeHatch({ includeMockHandoff: false });
 });
 
@@ -41375,11 +41419,13 @@ class RectifierPromptBuilder {
     const parts = [];
     const attemptWord = maxAttempts === 1 ? "1 attempt" : `${maxAttempts} attempts`;
     const exCount = story ? exceptionCountWord(story) : "three";
+    const prohibition = `Do NOT change test files or test behavior \u2014 see the ${exCount} narrow exceptions appended below.`;
+    const testDirective = story ? testEditHeadline(story, prohibition) : prohibition;
     parts.push(`Review failed after your implementation. Fix the following issues (${attemptWord} available before escalation):
 `);
     parts.push(renderPrioritizedFailures(failedChecks));
     parts.push(`
-Fix in priority order. After fixing each priority, re-run the failing check(s) at that level to verify they pass before moving on. Do NOT change test files or test behavior \u2014 see the ${exCount} narrow exceptions appended below. Commit your changes when all checks pass.`);
+Fix in priority order. After fixing each priority, re-run the failing check(s) at that level to verify they pass before moving on. ${testDirective} Commit your changes when all checks pass.`);
     parts.push(story ? escapeHatchFor(story) : CONTRADICTION_ESCAPE_HATCH);
     const guardrails = buildBehavioralGuardrailsSection("implementer", guardrailLevel ?? "lite");
     if (guardrails) {
@@ -41642,7 +41688,7 @@ ${testCommands}
 6. Ensure ALL tests pass before completing.
 
 **IMPORTANT:**
-- Do NOT modify test files \u2014 see the ${exceptionCountWord(story)} narrow exceptions appended below if you believe a test has a lint error, a PRD-contract mismatch, or belongs to a sibling story.
+- ${testEditHeadline(story, `Do NOT modify test files \u2014 see the ${exceptionCountWord(story)} narrow exceptions appended below if you believe a test has a lint error, a PRD-contract mismatch, or belongs to a sibling story.`)}
 - Do NOT loosen assertions to mask implementation bugs.
 - Focus on fixing the source code to meet the test requirements.
 - When running tests, run ONLY the failing test files shown above${cmd ? ` \u2014 NEVER run \`${cmd}\` without a file filter` : " \u2014 never run the full test suite without a file filter"}.
@@ -41742,7 +41788,7 @@ ${errors3}${reasoningSection}${historySection}
 2. Only fix findings that are actually valid problems
 3. Do NOT add keys, functions, or imports that already exist \u2014 check first
 
-Do NOT change test files or test behavior \u2014 see the ${exceptionCountWord(story)} narrow exceptions appended below.
+${testEditHeadline(story, `Do NOT change test files or test behavior \u2014 see the ${exceptionCountWord(story)} narrow exceptions appended below.`)}
 Do NOT add new features \u2014 only fix valid issues.
 Commit your fixes when done.${scopeConstraint}${escapeHatchFor(story)}`;
   }
@@ -41836,7 +41882,7 @@ Tests are failing. Fix the source so all tests pass \u2014 not just the ones lis
 4. Do not declare done until step 3 shows 0 failures.
 
 **IMPORTANT:**
-- Do NOT modify test files \u2014 see the ${exceptionCountWord(opts.story)} narrow exceptions appended below if you believe a test has a lint error, a PRD-contract mismatch, or belongs to a sibling story.
+- ${testEditHeadline(opts.story, `Do NOT modify test files \u2014 see the ${exceptionCountWord(opts.story)} narrow exceptions appended below if you believe a test has a lint error, a PRD-contract mismatch, or belongs to a sibling story.`)}
 - Do NOT loosen assertions to mask implementation bugs.
 - Focus on fixing the source code to meet the test requirements.`);
     parts.push(escapeHatchFor(opts.story));
@@ -53401,6 +53447,11 @@ function phasesToRevalidate(strategiesRun, allPhases) {
   }
   return allPhases.filter((p) => needed.has(p.kind));
 }
+function orderGateLast(phases) {
+  const rest = phases.filter((p) => p.kind !== "full-suite-gate");
+  const gates = phases.filter((p) => p.kind === "full-suite-gate");
+  return [...rest, ...gates];
+}
 function toReviewDecisionPayload(opName, output) {
   if (output === null || output === undefined || typeof output !== "object")
     return null;
@@ -53615,18 +53666,24 @@ async function runPhase(ctx, slot, phaseCosts, phaseOutputs, isThreeSession = fa
     scope.close();
   }
 }
-function withIncreasingFailuresBail(strategies, enabled) {
+function withIncreasingFailuresBail(strategies, enabled, consecutiveIncreases) {
   if (!enabled)
     return strategies;
+  const threshold = Math.max(1, consecutiveIncreases);
   return strategies.map((strategy) => ({
     ...strategy,
     bailWhen: (iterations) => {
       const userReason = strategy.bailWhen?.(iterations) ?? null;
       if (userReason !== null)
         return userReason;
-      const last = iterations[iterations.length - 1];
-      if (last && last.findingsAfter.length > last.findingsBefore.length) {
-        return `failure count increased: ${last.findingsBefore.length} -> ${last.findingsAfter.length}`;
+      if (iterations.length < threshold)
+        return null;
+      const trailing = iterations.slice(-threshold);
+      const allRegressed = trailing.every((it) => it.findingsAfter.length > it.findingsBefore.length);
+      if (allRegressed) {
+        const first = trailing[0];
+        const last = trailing[trailing.length - 1];
+        return `failure count increased for ${threshold} consecutive iteration(s): ${first.findingsBefore.length} -> ${last.findingsAfter.length}`;
       }
       return null;
     }
@@ -53656,13 +53713,14 @@ async function runRectification(ctx, state, phaseCosts, phaseOutputs) {
   const cycle = {
     findings: [...initialFindings],
     iterations: [],
-    strategies: withIncreasingFailuresBail(rectification.strategies, rectification.abortOnIncreasingFailures),
+    strategies: withIncreasingFailuresBail(rectification.strategies, rectification.abortOnIncreasingFailures, rectification.consecutiveIncreasesToBail ?? 1),
     config: { maxAttemptsTotal: rectification.maxAttempts, validatorRetries: 1 },
     validate: async (_validateCtx, opts) => {
       if (ctx.runtime.signal?.aborted)
         return { findings: [], shortCircuited: false };
       const lite = (opts?.mode ?? "full") === "lite";
-      const phases = phasesToRevalidate(opts?.strategiesRun, validationPhases);
+      const selected = phasesToRevalidate(opts?.strategiesRun, validationPhases);
+      const phases = lite ? orderGateLast(selected) : selected;
       getSafeLogger()?.debug("story-orchestrator", "rectification validate scope", {
         storyId: ctx.storyId,
         mode: opts?.mode ?? "full",
@@ -53672,9 +53730,6 @@ async function runRectification(ctx, state, phaseCosts, phaseOutputs) {
       const findings = [];
       let shortCircuited = false;
       for (const phase of phases) {
-        if (lite && phase.kind === "full-suite-gate") {
-          continue;
-        }
         await runPhase(ctx, phase.slot, phaseCosts, phaseOutputs);
         if (shouldSkipPhaseForRectification(phase, state, phaseOutputs))
           continue;
@@ -53990,9 +54045,6 @@ var init_story_orchestrator = __esm(() => {
 
 // src/execution/build-plan-for-strategy.ts
 import { join as join46 } from "path";
-function isThreeSessionStrategy(strategy) {
-  return THREE_SESSION_STRATEGIES2.has(strategy);
-}
 function requiresInitialRefCapture(strategy) {
   return isThreeSessionStrategy(strategy);
 }
@@ -54051,8 +54103,12 @@ async function buildPlanForStrategy(ctx, story, config2, testStrategy, inputs) {
       strategies.push(makeFullSuiteRectifyStrategy(story, config2));
     }
     if (config2.quality.autofix?.enabled !== false) {
-      strategies.push(makeAutofixImplementerStrategy(story, config2, sink));
-      strategies.push(makeAutofixTestWriterStrategy(story, config2, sink));
+      strategies.push(makeAutofixImplementerStrategy(story, config2, sink, {
+        includeAdversarialReview: !isThreeSession
+      }));
+      if (isThreeSession) {
+        strategies.push(makeAutofixTestWriterStrategy(story, config2, sink));
+      }
     }
     const postValidate = async (findings, _validateCtx) => {
       if (sink.testEdits.length === 0 && sink.mockHandoffs.length === 0)
@@ -54078,14 +54134,13 @@ async function buildPlanForStrategy(ctx, story, config2, testStrategy, inputs) {
   }
   return builder.build(ctx, { isThreeSession });
 }
-var THREE_SESSION_STRATEGIES2;
 var init_build_plan_for_strategy = __esm(() => {
+  init_config();
   init_operations();
   init_execution_gates();
   init_full_suite_rectify();
   init_test_runners();
   init_story_orchestrator();
-  THREE_SESSION_STRATEGIES2 = new Set(["three-session-tdd", "three-session-tdd-lite"]);
 });
 
 // src/execution/plan-inputs.ts
@@ -54285,7 +54340,8 @@ async function assemblePlanInputsFromCtx(ctx) {
   const rectificationInput = ctx.config.execution?.rectification?.enabled === true ? {
     maxAttempts: ctx.config.execution.rectification.maxAttemptsTotal,
     strategies: [],
-    abortOnIncreasingFailures: ctx.config.execution.rectification.abortOnIncreasingFailures
+    abortOnIncreasingFailures: ctx.config.execution.rectification.abortOnIncreasingFailures,
+    consecutiveIncreasesToBail: ctx.config.execution.rectification.consecutiveIncreasesToBail
   } : undefined;
   return {
     story,
@@ -54304,12 +54360,12 @@ async function assemblePlanInputsFromCtx(ctx) {
   };
 }
 var init_plan_inputs = __esm(() => {
+  init_config();
   init_context();
   init_errors();
   init_prompts();
   init_review();
   init_resolver();
-  init_build_plan_for_strategy();
 });
 
 // src/pipeline/stages/execution-helpers.ts
@@ -54836,6 +54892,7 @@ var init_post_run = __esm(() => {
 // src/pipeline/stages/execution.ts
 var RUNTIME_CRASH_CODES, executionStage, _executionDeps;
 var init_execution = __esm(() => {
+  init_config();
   init_agents();
   init_errors();
   init_build_plan_for_strategy();
@@ -58354,7 +58411,7 @@ var package_default;
 var init_package = __esm(() => {
   package_default = {
     name: "@nathapp/nax",
-    version: "0.68.3",
+    version: "0.68.5",
     description: "AI Coding Agent Orchestrator \u2014 loops until done",
     type: "module",
     bin: {
@@ -58449,8 +58506,8 @@ var init_version = __esm(() => {
   NAX_VERSION = package_default.version;
   NAX_COMMIT = (() => {
     try {
-      if (/^[0-9a-f]{6,10}$/.test("a1007103"))
-        return "a1007103";
+      if (/^[0-9a-f]{6,10}$/.test("1056817e"))
+        return "1056817e";
     } catch {}
     try {
       const result = Bun.spawnSync(["git", "rev-parse", "--short", "HEAD"], {
@@ -96908,6 +96965,7 @@ var FIELD_DESCRIPTIONS = {
   "execution.rectification.fullSuiteTimeoutSeconds": "Timeout for full test suite run in seconds",
   "execution.rectification.maxFailureSummaryChars": "Max characters in failure summary",
   "execution.rectification.abortOnIncreasingFailures": "Abort if failure count increases",
+  "execution.rectification.consecutiveIncreasesToBail": "Consecutive regressing iterations required before abortOnIncreasingFailures bails (default: 2; 1 = legacy behaviour)",
   "execution.rectification.escalateOnExhaustion": "Enable model tier escalation when attempts are exhausted with remaining failures",
   "execution.rectification.rethinkAtAttempt": "Attempt number at which 'rethink your approach' language is injected into the prompt (default: 2)",
   "execution.rectification.urgencyAtAttempt": "Attempt number at which 'final chance before escalation' urgency is added (default: 3)",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nathapp/nax",
-  "version": "0.68.3",
+  "version": "0.68.5",
   "description": "AI Coding Agent Orchestrator — loops until done",
   "type": "module",
   "bin": {