@nathapp/nax 0.44.0 → 0.46.0

package/dist/nax.js

@@ -3240,61 +3240,56 @@ async function withProcessTimeout(proc, timeoutMs, opts) {
   return { exitCode, timedOut };
 }
 
-// src/agents/types.ts
-var CompleteError;
-var init_types2 = __esm(() => {
-  CompleteError = class CompleteError extends Error {
-    exitCode;
-    constructor(message, exitCode) {
-      super(message);
-      this.exitCode = exitCode;
-      this.name = "CompleteError";
-    }
-  };
-});
-
-// src/agents/claude-complete.ts
-async function executeComplete(binary, prompt, options) {
-  const cmd = [binary, "-p", prompt];
-  if (options?.model) {
-    cmd.push("--model", options.model);
-  }
-  if (options?.jsonMode) {
-    cmd.push("--output-format", "json");
-  }
-  const { skipPermissions } = resolvePermissions(options?.config, "complete");
-  if (skipPermissions) {
-    cmd.push("--dangerously-skip-permissions");
-  }
-  const spawnOpts = { stdout: "pipe", stderr: "pipe" };
-  if (options?.workdir)
-    spawnOpts.cwd = options.workdir;
-  const proc = _completeDeps.spawn(cmd, spawnOpts);
-  const exitCode = await proc.exited;
-  const stdout = await new Response(proc.stdout).text();
-  const stderr = await new Response(proc.stderr).text();
-  const trimmed = stdout.trim();
-  if (exitCode !== 0) {
-    const errorDetails = stderr.trim() || trimmed;
-    const errorMessage = errorDetails || `complete() failed with exit code ${exitCode}`;
-    throw new CompleteError(errorMessage, exitCode);
-  }
-  if (!trimmed) {
-    throw new CompleteError("complete() returned empty output");
-  }
-  return trimmed;
+// src/config/test-strategy.ts
+function resolveTestStrategy(raw) {
+  if (!raw)
+    return "test-after";
+  if (VALID_TEST_STRATEGIES.includes(raw))
+    return raw;
+  if (raw === "tdd")
+    return "tdd-simple";
+  if (raw === "three-session")
+    return "three-session-tdd";
+  if (raw === "tdd-lite")
+    return "three-session-tdd-lite";
+  return "test-after";
 }
-var _completeDeps;
-var init_claude_complete = __esm(() => {
-  init_types2();
-  _completeDeps = {
-    spawn(cmd, opts) {
-      return Bun.spawn(cmd, opts);
-    }
-  };
+var VALID_TEST_STRATEGIES, COMPLEXITY_GUIDE = `## Complexity Classification Guide
+
+- simple: \u226450 LOC, single-file change, purely additive, no new dependencies \u2192 test-after
+- medium: 50\u2013200 LOC, 2\u20135 files, standard patterns, clear requirements \u2192 tdd-simple
+- complex: 200\u2013500 LOC, multiple modules, new abstractions or integrations \u2192 three-session-tdd
+- expert: 500+ LOC, architectural changes, cross-cutting concerns, high risk \u2192 three-session-tdd-lite
+
+### Security Override
+
+Security-critical functions (authentication, cryptography, tokens, sessions, credentials,
+password hashing, access control) must be classified at MINIMUM "medium" complexity
+regardless of LOC count. These require at minimum "tdd-simple" test strategy.`, TEST_STRATEGY_GUIDE = `## Test Strategy Guide
+
+- test-after: Simple changes with well-understood behavior. Write tests after implementation.
+- tdd-simple: Medium complexity. Write key tests first, implement, then fill coverage.
+- three-session-tdd: Complex stories. Full TDD cycle with separate test-writer and implementer sessions.
+- three-session-tdd-lite: Expert/high-risk stories. Full TDD with additional verifier session.`, GROUPING_RULES = `## Grouping Rules
+
+- Combine small, related tasks into a single "simple" or "medium" story.
+- Do NOT create separate stories for every single file or function unless complex.
+- Do NOT create standalone stories purely for test coverage or testing.
+  Each story's testStrategy already handles testing (tdd-simple writes tests first,
+  three-session-tdd uses separate test-writer session, test-after writes tests after).
+  Only create a dedicated test story for unique integration/E2E test logic that spans
+  multiple stories and cannot be covered by individual story test strategies.
+- Aim for coherent units of value. Maximum recommended stories: 10-15 per feature.`;
+var init_test_strategy = __esm(() => {
+  VALID_TEST_STRATEGIES = [
+    "test-after",
+    "tdd-simple",
+    "three-session-tdd",
+    "three-session-tdd-lite"
+  ];
 });
 
-// src/agents/claude-decompose.ts
+// src/agents/shared/decompose.ts
 function buildDecomposePrompt(options) {
   return `You are a requirements analyst. Break down the following feature specification into user stories and classify each story's complexity.
 
@@ -3316,24 +3311,13 @@ Decompose this spec into user stories. For each story, provide:
 9. reasoning: Why this complexity level
 10. estimatedLOC: Estimated lines of code to change
 11. risks: Array of implementation risks
-12. testStrategy: "three-session-tdd" | "test-after"
+12. testStrategy: "test-after" | "tdd-simple" | "three-session-tdd" | "three-session-tdd-lite"
 
-testStrategy rules:
-- "three-session-tdd": ONLY for complex/expert tasks that are security-critical (auth, encryption, tokens, credentials) or define public API contracts consumers depend on
-- "test-after": for all other tasks including simple/medium complexity
-- A "simple" complexity task should almost never be "three-session-tdd"
+${COMPLEXITY_GUIDE}
 
-Complexity classification rules:
-- simple: 1-3 files, <100 LOC, straightforward implementation, existing patterns
-- medium: 3-6 files, 100-300 LOC, moderate logic, some new patterns
-- complex: 6+ files, 300-800 LOC, architectural changes, cross-cutting concerns
-- expert: Security/crypto/real-time/distributed systems, >800 LOC, new infrastructure
+${TEST_STRATEGY_GUIDE}
 
-Grouping Guidelines:
-- Combine small, related tasks (e.g., multiple utility functions, interfaces) into a single "simple" or "medium" story.
-- Do NOT create separate stories for every single file or function unless complex.
-- Aim for coherent units of value (e.g., "Implement User Authentication" vs "Create User Interface", "Create Login Service").
-- Maximum recommended stories: 10-15 per feature. Group aggressively if list grows too long.
+${GROUPING_RULES}
 
 Consider:
 1. Does infrastructure exist? (e.g., "add caching" when no cache layer exists = complex)
@@ -3402,7 +3386,7 @@ ${output.slice(0, 500)}`);
       reasoning: String(record.reasoning || "No reasoning provided"),
       estimatedLOC: Number(record.estimatedLOC) || 0,
       risks: Array.isArray(record.risks) ? record.risks : [],
-      testStrategy: record.testStrategy === "three-session-tdd" ? "three-session-tdd" : record.testStrategy === "test-after" ? "test-after" : undefined
+      testStrategy: resolveTestStrategy(typeof record.testStrategy === "string" ? record.testStrategy : undefined)
     };
   });
   if (stories.length === 0) {
@@ -3416,8 +3400,65 @@ function coerceComplexity(value) {
   }
   return "medium";
 }
+var init_decompose = __esm(() => {
+  init_test_strategy();
+});
 
-// src/agents/cost.ts
+// src/agents/types.ts
+var CompleteError;
+var init_types2 = __esm(() => {
+  CompleteError = class CompleteError extends Error {
+    exitCode;
+    constructor(message, exitCode) {
+      super(message);
+      this.exitCode = exitCode;
+      this.name = "CompleteError";
+    }
+  };
+});
+
+// src/agents/claude/complete.ts
+async function executeComplete(binary, prompt, options) {
+  const cmd = [binary, "-p", prompt];
+  if (options?.model) {
+    cmd.push("--model", options.model);
+  }
+  if (options?.jsonMode) {
+    cmd.push("--output-format", "json");
+  }
+  const { skipPermissions } = resolvePermissions(options?.config, "complete");
+  if (skipPermissions) {
+    cmd.push("--dangerously-skip-permissions");
+  }
+  const spawnOpts = { stdout: "pipe", stderr: "pipe" };
+  if (options?.workdir)
+    spawnOpts.cwd = options.workdir;
+  const proc = _completeDeps.spawn(cmd, spawnOpts);
+  const exitCode = await proc.exited;
+  const stdout = await new Response(proc.stdout).text();
+  const stderr = await new Response(proc.stderr).text();
+  const trimmed = stdout.trim();
+  if (exitCode !== 0) {
+    const errorDetails = stderr.trim() || trimmed;
+    const errorMessage = errorDetails || `complete() failed with exit code ${exitCode}`;
+    throw new CompleteError(errorMessage, exitCode);
+  }
+  if (!trimmed) {
+    throw new CompleteError("complete() returned empty output");
+  }
+  return trimmed;
+}
+var _completeDeps;
+var init_complete = __esm(() => {
+  init_types2();
+  _completeDeps = {
+    spawn(cmd, opts) {
+      return Bun.spawn(cmd, opts);
+    }
+  };
+});
+
+// src/agents/claude/cost.ts
 function parseTokenUsage(output) {
   try {
     const jsonMatch = output.match(/\{[^}]*"usage"\s*:\s*\{[^}]*"input_tokens"\s*:\s*(\d+)[^}]*"output_tokens"\s*:\s*(\d+)[^}]*\}[^}]*\}/);
@@ -3520,7 +3561,7 @@ var init_cost = __esm(() => {
   };
 });
 
-// src/agents/claude-execution.ts
+// src/agents/claude/execution.ts
 function buildCommand(binary, options) {
   const model = options.modelDef.model;
   const { skipPermissions } = resolvePermissions(options.config, options.pipelineStage ?? "run");
@@ -3622,7 +3663,7 @@ async function executeOnce(binary, options, pidRegistry) {
   };
 }
 var MAX_AGENT_OUTPUT_CHARS = 5000, MAX_AGENT_STDERR_CHARS = 1000, SIGKILL_GRACE_PERIOD_MS = 5000, _runOnceDeps;
-var init_claude_execution = __esm(() => {
+var init_execution = __esm(() => {
   init_logger2();
   init_cost();
   _runOnceDeps = {
@@ -3635,7 +3676,7 @@ var init_claude_execution = __esm(() => {
   };
 });
 
-// src/agents/claude-interactive.ts
+// src/agents/claude/interactive.ts
 function runInteractiveMode(binary, options, pidRegistry) {
   const model = options.modelDef.model;
   const cmd = [binary, "--model", model, options.prompt];
@@ -3674,9 +3715,9 @@ function runInteractiveMode(binary, options, pidRegistry) {
     pid: proc.pid
   };
 }
-var init_claude_interactive = __esm(() => {
+var init_interactive = __esm(() => {
   init_logger2();
-  init_claude_execution();
+  init_execution();
 });
 
 // src/config/schema-types.ts
@@ -18098,7 +18139,7 @@ var init_schema = __esm(() => {
   init_defaults();
 });
 
-// src/agents/model-resolution.ts
+// src/agents/shared/model-resolution.ts
 var exports_model_resolution = {};
 __export(exports_model_resolution, {
   resolveBalancedModelDef: () => resolveBalancedModelDef
@@ -18119,7 +18160,7 @@ var init_model_resolution = __esm(() => {
   init_schema();
 });
 
-// src/agents/claude-plan.ts
+// src/agents/claude/plan.ts
 import { mkdtempSync, rmSync } from "fs";
 import { tmpdir } from "os";
 import { join } from "path";
@@ -18238,12 +18279,12 @@ async function runPlan(binary, options, pidRegistry, buildAllowedEnv2) {
     }
   }
 }
-var init_claude_plan = __esm(() => {
+var init_plan = __esm(() => {
   init_logger2();
   init_model_resolution();
 });
 
-// src/agents/claude.ts
+// src/agents/claude/adapter.ts
 class ClaudeCodeAdapter {
   name = "claude";
   displayName = "Claude Code";
@@ -18394,13 +18435,14 @@ class ClaudeCodeAdapter {
   }
 }
 var _decomposeDeps, _claudeAdapterDeps;
-var init_claude = __esm(() => {
+var init_adapter = __esm(() => {
   init_pid_registry();
   init_logger2();
-  init_claude_complete();
-  init_claude_execution();
-  init_claude_interactive();
-  init_claude_plan();
+  init_decompose();
+  init_complete();
+  init_execution();
+  init_interactive();
+  init_plan();
   _decomposeDeps = {
     spawn(cmd, opts) {
       return Bun.spawn(cmd, opts);
@@ -18411,6 +18453,12 @@ var init_claude = __esm(() => {
   };
 });
 
+// src/agents/claude/index.ts
+var init_claude = __esm(() => {
+  init_adapter();
+  init_execution();
+});
+
 // src/utils/errors.ts
 function errorMessage(err) {
   return err instanceof Error ? err.message : String(err);
@@ -18935,11 +18983,40 @@ function parseAcpxJsonOutput(rawOutput) {
 `).filter((l) => l.trim());
   let text = "";
   let tokenUsage;
+  let exactCostUsd;
   let stopReason;
   let error48;
   for (const line of lines) {
     try {
       const event = JSON.parse(line);
+      if (event.jsonrpc === "2.0") {
+        if (event.method === "session/update" && event.params?.update) {
+          const update = event.params.update;
+          if (update.sessionUpdate === "agent_message_chunk" && update.content?.type === "text" && update.content.text) {
+            text += update.content.text;
+          }
+          if (update.sessionUpdate === "usage_update" && typeof update.cost?.amount === "number") {
+            exactCostUsd = update.cost.amount;
+          }
+        }
+        if (event.id !== undefined && event.result && typeof event.result === "object") {
+          const result = event.result;
+          if (result.stopReason)
+            stopReason = result.stopReason;
+          if (result.stop_reason)
+            stopReason = result.stop_reason;
+          if (result.usage && typeof result.usage === "object") {
+            const u = result.usage;
+            tokenUsage = {
+              input_tokens: u.inputTokens ?? u.input_tokens ?? 0,
+              output_tokens: u.outputTokens ?? u.output_tokens ?? 0,
+              cache_read_input_tokens: u.cachedReadTokens ?? u.cache_read_input_tokens ?? 0,
+              cache_creation_input_tokens: u.cachedWriteTokens ?? u.cache_creation_input_tokens ?? 0
+            };
+          }
+        }
+        continue;
+      }
       if (event.content && typeof event.content === "string")
         text += event.content;
       if (event.text && typeof event.text === "string")
@@ -18966,7 +19043,7 @@ function parseAcpxJsonOutput(rawOutput) {
         text = line;
     }
   }
-  return { text: text.trim(), tokenUsage, stopReason, error: error48 };
+  return { text: text.trim(), tokenUsage, exactCostUsd, stopReason, error: error48 };
 }
 
 // src/agents/acp/spawn-client.ts
@@ -19065,8 +19142,9 @@ class SpawnAcpSession {
         const parsed = parseAcpxJsonOutput(stdout);
         return {
           messages: [{ role: "assistant", content: parsed.text || "" }],
-          stopReason: "end_turn",
-          cumulative_token_usage: parsed.tokenUsage
+          stopReason: parsed.stopReason ?? "end_turn",
+          cumulative_token_usage: parsed.tokenUsage,
+          exactCostUsd: parsed.exactCostUsd
         };
       } catch (err) {
         getSafeLogger()?.warn("acp-adapter", "Failed to parse session prompt response", {
@@ -19527,7 +19605,13 @@ class AcpAgentAdapter {
     let lastResponse = null;
     let timedOut = false;
     const runState = { succeeded: false };
-    const totalTokenUsage = { input_tokens: 0, output_tokens: 0 };
+    const totalTokenUsage = {
+      input_tokens: 0,
+      output_tokens: 0,
+      cache_read_input_tokens: 0,
+      cache_creation_input_tokens: 0
+    };
+    let totalExactCostUsd;
     try {
       let currentPrompt = options.prompt;
       let turnCount = 0;
@@ -19546,6 +19630,11 @@ class AcpAgentAdapter {
         if (lastResponse.cumulative_token_usage) {
           totalTokenUsage.input_tokens += lastResponse.cumulative_token_usage.input_tokens ?? 0;
           totalTokenUsage.output_tokens += lastResponse.cumulative_token_usage.output_tokens ?? 0;
+          totalTokenUsage.cache_read_input_tokens += lastResponse.cumulative_token_usage.cache_read_input_tokens ?? 0;
+          totalTokenUsage.cache_creation_input_tokens += lastResponse.cumulative_token_usage.cache_creation_input_tokens ?? 0;
+        }
+        if (lastResponse.exactCostUsd !== undefined) {
+          totalExactCostUsd = (totalExactCostUsd ?? 0) + lastResponse.exactCostUsd;
         }
         const outputText = extractOutput(lastResponse);
         const question = extractQuestion(outputText);
@@ -19592,7 +19681,7 @@ class AcpAgentAdapter {
     }
     const success2 = lastResponse?.stopReason === "end_turn";
     const output = extractOutput(lastResponse);
-    const estimatedCost = totalTokenUsage.input_tokens > 0 || totalTokenUsage.output_tokens > 0 ? estimateCostFromTokenUsage(totalTokenUsage, options.modelDef.model) : 0;
+    const estimatedCost = totalExactCostUsd ?? (totalTokenUsage.input_tokens > 0 || totalTokenUsage.output_tokens > 0 ? estimateCostFromTokenUsage(totalTokenUsage, options.modelDef.model) : 0);
     return {
       success: success2,
       exitCode: success2 ? 0 : 1,
@@ -19642,6 +19731,12 @@ class AcpAgentAdapter {
         if (!unwrapped) {
           throw new CompleteError("complete() returned empty output");
         }
+        if (response.exactCostUsd !== undefined) {
+          getSafeLogger()?.info("acp-adapter", "complete() cost", {
+            costUsd: response.exactCostUsd,
+            model
+          });
+        }
         return unwrapped;
       } catch (err) {
         const error48 = err instanceof Error ? err : new Error(String(err));
@@ -19728,8 +19823,9 @@ class AcpAgentAdapter {
   }
 }
 var MAX_AGENT_OUTPUT_CHARS2 = 5000, MAX_RATE_LIMIT_RETRIES = 3, INTERACTION_TIMEOUT_MS, AGENT_REGISTRY, DEFAULT_ENTRY, _acpAdapterDeps, MAX_SESSION_AGE_MS;
-var init_adapter = __esm(() => {
+var init_adapter2 = __esm(() => {
   init_logger2();
+  init_decompose();
   init_spawn_client();
   init_types2();
   init_cost2();
@@ -19774,7 +19870,7 @@ var init_adapter = __esm(() => {
   MAX_SESSION_AGE_MS = 2 * 60 * 60 * 1000;
 });
 
-// src/agents/adapters/aider.ts
+// src/agents/aider/adapter.ts
 class AiderAdapter {
   name = "aider";
   displayName = "Aider";
@@ -19839,7 +19935,7 @@ class AiderAdapter {
   }
 }
 var _aiderCompleteDeps, MAX_AGENT_OUTPUT_CHARS3 = 5000;
-var init_aider = __esm(() => {
+var init_adapter3 = __esm(() => {
   init_types2();
   _aiderCompleteDeps = {
     which(name) {
@@ -19851,7 +19947,7 @@ var init_aider = __esm(() => {
   };
 });
 
-// src/agents/adapters/codex.ts
+// src/agents/codex/adapter.ts
 class CodexAdapter {
   name = "codex";
   displayName = "Codex";
@@ -19914,7 +20010,7 @@ class CodexAdapter {
   }
 }
 var _codexRunDeps, _codexCompleteDeps, MAX_AGENT_OUTPUT_CHARS4 = 5000;
-var init_codex = __esm(() => {
+var init_adapter4 = __esm(() => {
   init_types2();
   _codexRunDeps = {
     which(name) {
@@ -19931,7 +20027,7 @@ var init_codex = __esm(() => {
   };
 });
 
-// src/agents/adapters/gemini.ts
+// src/agents/gemini/adapter.ts
 class GeminiAdapter {
   name = "gemini";
   displayName = "Gemini CLI";
@@ -20014,7 +20110,7 @@ class GeminiAdapter {
   }
 }
 var _geminiRunDeps, _geminiCompleteDeps, MAX_AGENT_OUTPUT_CHARS5 = 5000;
-var init_gemini = __esm(() => {
+var init_adapter5 = __esm(() => {
   init_types2();
   _geminiRunDeps = {
     which(name) {
@@ -20031,7 +20127,7 @@ var init_gemini = __esm(() => {
   };
 });
 
-// src/agents/adapters/opencode.ts
+// src/agents/opencode/adapter.ts
 class OpenCodeAdapter {
   name = "opencode";
   displayName = "OpenCode";
@@ -20076,7 +20172,7 @@ class OpenCodeAdapter {
   }
 }
 var _opencodeCompleteDeps;
-var init_opencode = __esm(() => {
+var init_adapter6 = __esm(() => {
   init_types2();
   _opencodeCompleteDeps = {
     which(name) {
@@ -20168,12 +20264,12 @@ function createAgentRegistry(config2) {
 var ALL_AGENTS;
 var init_registry = __esm(() => {
   init_logger2();
+  init_adapter2();
+  init_adapter3();
   init_adapter();
-  init_aider();
-  init_codex();
-  init_gemini();
-  init_opencode();
-  init_claude();
+  init_adapter4();
+  init_adapter5();
+  init_adapter6();
   ALL_AGENTS = [
     new ClaudeCodeAdapter,
     new CodexAdapter,
@@ -22042,7 +22138,7 @@ var package_default;
 var init_package = __esm(() => {
   package_default = {
     name: "@nathapp/nax",
-    version: "0.44.0",
+    version: "0.46.0",
     description: "AI Coding Agent Orchestrator \u2014 loops until done",
     type: "module",
     bin: {
@@ -22115,8 +22211,8 @@ var init_version = __esm(() => {
   NAX_VERSION = package_default.version;
   NAX_COMMIT = (() => {
     try {
-      if (/^[0-9a-f]{6,10}$/.test("05b2442"))
-        return "05b2442";
+      if (/^[0-9a-f]{6,10}$/.test("6a485b9"))
+        return "6a485b9";
     } catch {}
     try {
       const result = Bun.spawnSync(["git", "rev-parse", "--short", "HEAD"], {
@@ -23716,6 +23812,20 @@ var init_runner = __esm(() => {
   init_logger2();
 });
 
+// src/utils/log-test-output.ts
+function logTestOutput(logger, stage, output, opts = {}) {
+  if (!logger || !output)
+    return;
+  const tailLines = opts.tailLines ?? 20;
+  const lines = output.split(`
+`).slice(-tailLines).join(`
+`);
+  logger.debug(stage, "Test output (tail)", {
+    ...opts.storyId !== undefined && { storyId: opts.storyId },
+    output: lines
+  });
+}
+
 // src/pipeline/stages/acceptance.ts
 var exports_acceptance = {};
 __export(exports_acceptance, {
@@ -23799,10 +23909,8 @@ ${stderr}`;
         return { action: "continue" };
       }
       if (failedACs.length === 0 && exitCode !== 0) {
-        logger.error("acceptance", "Tests errored with no AC failures parsed", {
-          exitCode,
-          output
-        });
+        logger.error("acceptance", "Tests errored with no AC failures parsed", { exitCode });
+        logTestOutput(logger, "acceptance", output);
         ctx.acceptanceFailures = {
           failedACs: ["AC-ERROR"],
           testOutput: output
@@ -23820,10 +23928,8 @@ ${stderr}`;
             overrides: overriddenFailures.map((acId) => ({ acId, reason: overrides[acId] }))
           });
         }
-        logger.error("acceptance", "Acceptance tests failed", {
-          failedACs: actualFailures,
-          output
-        });
+        logger.error("acceptance", "Acceptance tests failed", { failedACs: actualFailures });
+        logTestOutput(logger, "acceptance", output);
         ctx.acceptanceFailures = {
           failedACs: actualFailures,
           testOutput: output
@@ -25394,6 +25500,11 @@ async function buildStoryContextFull(prd, story, config2) {
 }
 function getAllReadyStories(prd) {
   const completedIds = new Set(prd.userStories.filter((s) => s.passes || s.status === "skipped").map((s) => s.id));
+  const logger = getSafeLogger2();
+  logger?.debug("routing", "getAllReadyStories: completed set", {
+    completedIds: [...completedIds],
+    totalStories: prd.userStories.length
+  });
   return prd.userStories.filter((s) => !s.passes && s.status !== "skipped" && s.status !== "failed" && s.status !== "paused" && s.status !== "blocked" && s.dependencies.every((dep) => completedIds.has(dep)));
 }
 var CONTEXT_MAX_TOKENS = 1e5, CONTEXT_RESERVED_TOKENS = 1e4;
@@ -25567,7 +25678,7 @@ ${pluginMarkdown}` : pluginMarkdown;
   };
 });
 
-// src/agents/validation.ts
+// src/agents/shared/validation.ts
 function validateAgentForTier(agent, tier) {
   return agent.capabilities.supportedTiers.includes(tier);
 }
@@ -25581,7 +25692,7 @@ function describeAgentCapabilities(agent) {
   return `${agent.name}: tiers=[${tiers}], maxTokens=${maxTokens}, features=[${features}]`;
 }
 
-// src/agents/version-detection.ts
+// src/agents/shared/version-detection.ts
 async function getAgentVersion(binaryName) {
   try {
     const proc = _versionDetectionDeps.spawn([binaryName, "--version"], {
@@ -27694,7 +27805,7 @@ function routeTddFailure(failureCategory, isLiteMode, ctx, reviewReason) {
   };
 }
 var executionStage, _executionDeps;
-var init_execution = __esm(() => {
+var init_execution2 = __esm(() => {
   init_agents();
   init_config();
   init_triggers();
@@ -29049,6 +29160,7 @@ var init_regression2 = __esm(() => {
         storyId: ctx.story.id,
         failCount: result.failCount
       });
+      logTestOutput(logger, "regression", result.rawOutput, { storyId: ctx.story.id });
       pipelineEventBus.emit({
         type: "regression:detected",
         storyId: ctx.story.id,
@@ -29347,16 +29459,8 @@ var init_verify = __esm(() => {
             storyId: ctx.story.id
           });
         }
-        if (result.output && result.status !== "TIMEOUT") {
-          const outputLines = result.output.split(`
-`).slice(-20);
-          if (outputLines.length > 0) {
-            logger.debug("verify", "Test output preview", {
-              storyId: ctx.story.id,
-              output: outputLines.join(`
-`)
-            });
-          }
+        if (result.status !== "TIMEOUT") {
+          logTestOutput(logger, "verify", result.output, { storyId: ctx.story.id });
         }
         return {
           action: "escalate",
@@ -29401,7 +29505,7 @@ var init_stages = __esm(() => {
   init_completion();
   init_constitution2();
   init_context2();
-  init_execution();
+  init_execution2();
   init_optimizer2();
   init_prompt();
   init_queue_check();
@@ -29416,7 +29520,7 @@ var init_stages = __esm(() => {
   init_context2();
   init_prompt();
   init_optimizer2();
-  init_execution();
+  init_execution2();
   init_verify();
   init_rectify();
   init_review();
@@ -31237,7 +31341,8 @@ async function executeFixStory(ctx, story, prd, iterations) {
     featureDir: ctx.featureDir,
     hooks: ctx.hooks,
     plugins: ctx.pluginRegistry,
-    storyStartTime: new Date().toISOString()
+    storyStartTime: new Date().toISOString(),
+    agentGetFn: ctx.agentGetFn
   };
   const result = await runPipeline(defaultPipeline, fixContext, ctx.eventEmitter);
   logger?.info("acceptance", `Fix story ${story.id} ${result.success ? "passed" : "failed"}`);
@@ -31273,7 +31378,8 @@ async function runAcceptanceLoop(ctx) {
       workdir: ctx.workdir,
       featureDir: ctx.featureDir,
       hooks: ctx.hooks,
-      plugins: ctx.pluginRegistry
+      plugins: ctx.pluginRegistry,
+      agentGetFn: ctx.agentGetFn
     };
     const { acceptanceStage: acceptanceStage2 } = await Promise.resolve().then(() => (init_acceptance2(), exports_acceptance));
     const acceptanceResult = await acceptanceStage2.execute(acceptanceContext);
@@ -32254,7 +32360,7 @@ function resolveMaxConcurrency(parallel) {
   }
   return Math.max(1, parallel);
 }
-async function executeParallel(stories, prdPath, projectRoot, config2, hooks, plugins, prd, featureDir, parallel, eventEmitter) {
+async function executeParallel(stories, prdPath, projectRoot, config2, hooks, plugins, prd, featureDir, parallel, eventEmitter, agentGetFn) {
   const logger = getSafeLogger();
   const maxConcurrency = resolveMaxConcurrency(parallel);
   const worktreeManager = new WorktreeManager;
@@ -32284,7 +32390,8 @@ async function executeParallel(stories, prdPath, projectRoot, config2, hooks, pl
       featureDir,
       hooks,
       plugins,
-      storyStartTime: new Date().toISOString()
+      storyStartTime: new Date().toISOString(),
+      agentGetFn
     };
     const worktreePaths = new Map;
     for (const story of batch) {
@@ -32658,7 +32765,7 @@ async function runParallelExecution(options, initialPrd) {
   const batchStoryMetrics = [];
   let conflictedStories = [];
   try {
-    const parallelResult = await _parallelExecutorDeps.executeParallel(readyStories, prdPath, workdir, config2, hooks, pluginRegistry, prd, featureDir, parallelCount, eventEmitter);
+    const parallelResult = await _parallelExecutorDeps.executeParallel(readyStories, prdPath, workdir, config2, hooks, pluginRegistry, prd, featureDir, parallelCount, eventEmitter, options.agentGetFn);
     const batchDurationMs = Date.now() - batchStartMs;
     const batchCompletedAt = new Date().toISOString();
     prd = parallelResult.updatedPrd;
@@ -34403,7 +34510,7 @@ async function setupRun(options) {
   } else {
     logger?.warn("precheck", "Precheck validations skipped (--skip-precheck)");
   }
-  const { sweepStaleFeatureSessions: sweepStaleFeatureSessions2 } = await Promise.resolve().then(() => (init_adapter(), exports_adapter));
+  const { sweepStaleFeatureSessions: sweepStaleFeatureSessions2 } = await Promise.resolve().then(() => (init_adapter2(), exports_adapter));
   await sweepStaleFeatureSessions2(workdir, feature).catch(() => {});
   const lockAcquired = await acquireLock(workdir);
   if (!lockAcquired) {
@@ -65854,17 +65961,13 @@ init_registry();
 import { existsSync as existsSync9 } from "fs";
 import { join as join10 } from "path";
 import { createInterface } from "readline";
+init_test_strategy();
 init_pid_registry();
 init_logger2();
 
 // src/prd/schema.ts
+init_test_strategy();
 var VALID_COMPLEXITY = ["simple", "medium", "complex", "expert"];
-var VALID_TEST_STRATEGIES = [
-  "test-after",
-  "tdd-simple",
-  "three-session-tdd",
-  "three-session-tdd-lite"
-];
 var STORY_ID_NO_SEPARATOR = /^([A-Za-z]+)(\d+)$/;
 function extractJsonFromMarkdown(text) {
   const match = text.match(/```(?:json)?\s*\n([\s\S]*?)\n?\s*```/);
@@ -65934,9 +66037,7 @@ function validateStory(raw, index, allIds) {
     throw new Error(`[schema] story[${index}].routing.complexity "${rawComplexity}" is invalid. Valid values: ${VALID_COMPLEXITY.join(", ")}`);
   }
   const rawTestStrategy = routing.testStrategy ?? s.testStrategy;
-  const STRATEGY_ALIASES = { "tdd-lite": "three-session-tdd-lite" };
-  const normalizedStrategy = typeof rawTestStrategy === "string" ? STRATEGY_ALIASES[rawTestStrategy] ?? rawTestStrategy : rawTestStrategy;
-  const testStrategy = normalizedStrategy !== undefined && VALID_TEST_STRATEGIES.includes(normalizedStrategy) ? normalizedStrategy : "tdd-simple";
+  const testStrategy = resolveTestStrategy(typeof rawTestStrategy === "string" ? rawTestStrategy : undefined);
   const rawDeps = s.dependencies;
   const dependencies = Array.isArray(rawDeps) ? rawDeps : [];
   for (const dep of dependencies) {
@@ -66203,19 +66304,11 @@ Generate a JSON object with this exact structure (no markdown, no explanation \u
   ]
 }
 
-## Complexity Classification Guide
+${COMPLEXITY_GUIDE}
 
-- simple: \u226450 LOC, single-file change, purely additive, no new dependencies \u2192 test-after
-- medium: 50\u2013200 LOC, 2\u20135 files, standard patterns, clear requirements \u2192 tdd-simple
-- complex: 200\u2013500 LOC, multiple modules, new abstractions or integrations \u2192 three-session-tdd
-- expert: 500+ LOC, architectural changes, cross-cutting concerns, high risk \u2192 three-session-tdd-lite
+${TEST_STRATEGY_GUIDE}
 
-## Test Strategy Guide
-
-- test-after: Simple changes with well-understood behavior. Write tests after implementation.
-- tdd-simple: Medium complexity. Write key tests first, implement, then fill coverage.
-- three-session-tdd: Complex stories. Full TDD cycle with separate test-writer and implementer sessions.
-- three-session-tdd-lite: Expert/high-risk stories. Full TDD with additional verifier session.
+${GROUPING_RULES}
 
 ${outputFilePath ? `Write the PRD JSON directly to this file path: ${outputFilePath}
 Do NOT output the JSON to the conversation. Write the file, then reply with a brief confirmation.` : "Output ONLY the JSON object. Do not wrap in markdown code blocks."}`;
@@ -69051,7 +69144,7 @@ async function unlockCommand(options) {
 init_config();
 
 // src/execution/runner.ts
-init_adapter();
+init_adapter2();
 init_registry();
 init_hooks();
 init_logger2();
@@ -69227,9 +69320,20 @@ async function runExecutionPhase(options, prd, pluginRegistry) {
     batchingEnabled: options.useBatch
   });
   clearCache();
-  const batchPlan = options.useBatch ? precomputeBatchPlan(getAllReadyStories(prd), 4) : [];
+  const readyStories = getAllReadyStories(prd);
+  logger?.debug("routing", "Ready stories for batch routing", {
+    readyCount: readyStories.length,
+    readyIds: readyStories.map((s) => s.id),
+    allStories: prd.userStories.map((s) => ({
+      id: s.id,
+      status: s.status,
+      passes: s.passes,
+      deps: s.dependencies
+    }))
+  });
+  const batchPlan = options.useBatch ? precomputeBatchPlan(readyStories, 4) : [];
   if (options.useBatch) {
-    await tryLlmBatchRoute(options.config, getAllReadyStories(prd), "routing");
+    await tryLlmBatchRoute(options.config, readyStories, "routing");
   }
   if (options.parallel !== undefined) {
     const runParallelExecution2 = options.runParallelExecution ?? (await Promise.resolve().then(() => (init_parallel_executor(), exports_parallel_executor))).runParallelExecution;
@@ -76991,9 +77095,10 @@ program2.command("run").description("Run the orchestration loop for a feature").
       }
     }
     try {
-      mkdirSync6(featureDir, { recursive: true });
+      const planLogDir = join43(featureDir, "plan");
+      mkdirSync6(planLogDir, { recursive: true });
       const planLogId = new Date().toISOString().replace(/:/g, "-").replace(/\..+/, "");
-      const planLogPath = join43(featureDir, `plan-${planLogId}.jsonl`);
+      const planLogPath = join43(planLogDir, `${planLogId}.jsonl`);
       initLogger({ level: "info", filePath: planLogPath, useChalk: false, headless: true });
       console.log(source_default.dim(`   [Plan log: ${planLogPath}]`));
       console.log(source_default.dim("   [Planning phase: generating PRD from spec]"));
@@ -77238,10 +77343,10 @@ Use: nax plan -f <feature> --from <spec>`));
     process.exit(1);
   }
   const config2 = await loadConfig(workdir);
-  const featureLogDir = join43(naxDir, "features", options.feature);
+  const featureLogDir = join43(naxDir, "features", options.feature, "plan");
   mkdirSync6(featureLogDir, { recursive: true });
   const planLogId = new Date().toISOString().replace(/:/g, "-").replace(/\..+/, "");
-  const planLogPath = join43(featureLogDir, `plan-${planLogId}.jsonl`);
+  const planLogPath = join43(featureLogDir, `${planLogId}.jsonl`);
   initLogger({ level: "info", filePath: planLogPath, useChalk: false, headless: true });
   console.log(source_default.dim(`   [Plan log: ${planLogPath}]`));
   try {