ralphctl 0.2.1 → 0.2.3

package/README.md

@@ -4,53 +4,74 @@
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue?style=flat&logo=opensourceinitiative&logoColor=white)](./LICENSE)
 [![TypeScript](https://img.shields.io/badge/TypeScript-5.9-3178c6?style=flat&logo=typescript&logoColor=white)](https://www.typescriptlang.org/)
 [![Node.js](https://img.shields.io/badge/node-%E2%89%A5_24-5fa04e?style=flat&logo=nodedotjs&logoColor=white)](https://nodejs.org/)
-[![code style: prettier](https://img.shields.io/badge/code_style-prettier-ff69b4?style=flat&logo=prettier&logoColor=white)](https://prettier.io/)
-[![ESLint](https://img.shields.io/badge/ESLint-4b32c3?style=flat&logo=eslint&logoColor=white)](https://eslint.org/)
 [![PRs Welcome](https://img.shields.io/badge/PRs-welcome-brightgreen?style=flat&logo=git&logoColor=white)](./CONTRIBUTING.md)
 [![Claude Code](https://img.shields.io/badge/Claude_Code-191919?style=flat&logo=anthropic&logoColor=white)](https://docs.anthropic.com/en/docs/claude-code)
 [![GitHub Copilot](https://img.shields.io/badge/GitHub_Copilot-000?style=flat&logo=githubcopilot&logoColor=white)](https://docs.github.com/en/copilot/github-copilot-in-the-cli)
-[![Built with Donuts](https://img.shields.io/badge/%F0%9F%8D%A9-Built_with_Donuts-ff6f00?style=flat)](https://github.com/lukas-grigis/ralphctl)
 
 ```
-  🍩 ██████╗  █████╗ ██╗     ██████╗ ██╗  ██╗ ██████╗████████╗██╗     🍩
-     ██╔══██╗██╔══██╗██║     ██╔══██╗██║  ██║██╔════╝╚══██╔══╝██║
-     ██████╔╝███████║██║     ██████╔╝███████║██║        ██║   ██║
-     ██╔══██╗██╔══██║██║     ██╔═══╝ ██╔══██║██║        ██║   ██║
-     ██║  ██║██║  ██║███████╗██║     ██║  ██║╚██████╗   ██║   ███████╗
-     ╚═╝  ╚═╝╚═╝  ╚═╝╚══════╝╚═╝     ╚═╝  ╚═╝ ╚═════╝   ╚═╝   ╚══════╝
+  ██████╗  █████╗ ██╗     ██████╗ ██╗  ██╗ ██████╗████████╗██╗
+  ██╔══██╗██╔══██╗██║     ██╔══██╗██║  ██║██╔════╝╚══██╔══╝██║
+  ██████╔╝███████║██║     ██████╔╝███████║██║        ██║   ██║
+  ██╔══██╗██╔══██║██║     ██╔═══╝ ██╔══██║██║        ██║   ██║
+  ██║  ██║██║  ██║███████╗██║     ██║  ██║╚██████╗   ██║   ███████╗
+  ╚═╝  ╚═╝╚═╝  ╚═╝╚══════╝╚═╝     ╚═╝  ╚═╝ ╚═════╝   ╚═╝   ╚══════╝
 ```
 
-**Agent harness for long-running AI coding tasks — orchestrates [Claude Code](https://docs.anthropic.com/en/docs/claude-code) & [GitHub Copilot](https://docs.github.com/en/copilot/github-copilot-in-the-cli) across repositories.**
+**Agent harness for long-running AI coding tasks —
+orchestrates [Claude Code](https://docs.anthropic.com/en/docs/claude-code) & [GitHub Copilot](https://docs.github.com/en/copilot/github-copilot-in-the-cli)
+across repositories.**
 
 > _"I'm helping!"_ — Ralph Wiggum
 
 > [!NOTE]
 > **Early access.** RalphCTL is under active development. Things work, but expect rough edges and breaking changes
-> before 1.0. Read the [blog post](https://lukasgrigis.dev/blog/building-ralphctl) for the backstory.
+> before 1.0.
 
-RalphCTL decomposes work into dependency-ordered tasks, executes them through AI coding agents, and runs a
-[generator-evaluator loop](https://www.anthropic.com/engineering/harness-design-long-running-apps) to catch issues
-before moving on. It manages context across sessions so nothing gets lost — whether you're working on a single ticket
-or coordinating changes across multiple repositories. Ralph Wiggum personality included because why not.
+---
+
+## Why ralphctl?
+
+AI coding agents are powerful but lose context on long tasks, need babysitting when things break, and have no way to
+coordinate changes across multiple repositories. RalphCTL decomposes your work into dependency-ordered tasks, runs each
+one through a [generator-evaluator loop](https://www.anthropic.com/engineering/harness-design-long-running-apps) that
+catches issues before moving on, and persists context across sessions so nothing gets lost. You describe what to build —
+ralphctl handles the rest.
 
 ---
 
-## Install
+## How It Works
 
-```bash
-npm install -g ralphctl
 ```
+  You describe what to build           ralphctl handles the rest
+  ─────────────────────────           ─────────────────────────────────
+  ┌──────────┐   ┌──────────┐        ┌────────┐   ┌──────┐   ┌─────────┐
+  │  Create  │──>│   Add    │───────>│ Refine │──>│ Plan │──>│ Execute │
+  │  Sprint  │   │ Tickets  │        │ (WHAT) │   │(HOW) │   │  Loop   │
+  └──────────┘   └──────────┘        └────────┘   └──────┘   └─────────┘
+                                          │            │           │
+                                     AI clarifies  AI generates  AI implements
+                                     requirements  task graph    + AI reviews
+                                     with you      from specs    each task
+```
+
+- **Dependency-ordered execution** — tasks run in the right sequence, one per repo at a time, with parallel execution
+  where possible
+- **Generator-evaluator cycle** — an independent AI reviewer checks each task against its spec; if it fails, the
+  generator gets feedback and iterates
+- **Context persistence** — sprint state, progress history, and task context survive across sessions; interrupted work
+  resumes where it left off
 
-This installs the `ralphctl` command globally.
+---
 
-### Prerequisites
+## Quick Start
 
-- [Node.js](https://nodejs.org/) **>= 24.0.0**
-- [Git](https://git-scm.com/)
-- Either [Claude CLI](https://docs.anthropic.com/en/docs/claude-code)
-  or [GitHub Copilot CLI](https://docs.github.com/en/copilot/github-copilot-in-the-cli) installed and authenticated
+```bash
+npm install -g ralphctl
+```
 
-### 2-Minute Quick Start
+Requires [Node.js](https://nodejs.org/) >= 24, [Git](https://git-scm.com/), and
+either [Claude CLI](https://docs.anthropic.com/en/docs/claude-code)
+or [GitHub Copilot CLI](https://docs.github.com/en/copilot/github-copilot-in-the-cli) installed and authenticated.
 
 ```bash
 # 1. Register a project (points to your repo)
@@ -68,36 +89,65 @@ ralphctl sprint plan
 ralphctl sprint start
 ```
 
-Or just run `ralphctl` with no arguments for an interactive menu that walks you through everything.
+Or run `ralphctl` with no arguments for an interactive menu that walks you through everything.
 
 ---
 
-## Table of Contents
+## Features
 
-- [Features](#features)
-- [CLI Overview](#cli-overview)
-- [AI Provider Configuration](#ai-provider-configuration)
-- [Documentation](#documentation)
-- [Development](#development)
-- [Contributing](#contributing)
-- [License](#license)
+- **Break big tickets into small tasks** — dependency-ordered so they execute in the right sequence
+- **Catch mistakes before they compound** — independent AI review after each task, iterating until quality passes or
+  budget is exhausted
+- **Coordinate across repositories** — one sprint can span multiple repos with automatic dependency tracking
+- **Run tasks in parallel** — one per repo, with rate-limit backoff and automatic session resume
+- **Separate the what from the how** — AI clarifies requirements first, then generates implementation tasks, with human
+  approval gates
+- **Pick up where you left off** — full state persistence across sessions; interrupted work resumes automatically
+- **Pair or let it run** — work alongside your AI agent interactively, or let it execute unattended
+- **Zero-memorization start** — run `ralphctl` with no args for a guided menu
 
 ---
 
-## Features
+## Configuration
+
+RalphCTL supports **Claude Code** and **GitHub Copilot** as AI backends.
 
-- **Task decomposition** — breaks tickets into dependency-ordered tasks with topological sort
-- **Generator-evaluator loop** — independent AI review after each task; iterates until quality passes or budget exhausted
-- **Multi-repo orchestration** — coordinate changes across multiple repositories in a single run
-- **Parallel execution** — one task per repo at a time, with automatic rate limit backoff and session resume
-- **Two-phase planning** — clarify requirements first (what), then generate tasks (how), with a human approval gate
-- **Context persistence** — state survives across sessions; interrupted work resumes where it left off
-- **Interactive or headless** — pair with your AI agent in a session, or let it run unattended
-- **Menu mode** — run `ralphctl` with no arguments for an interactive menu
+```bash
+ralphctl config set provider claude      # Use Claude Code
+ralphctl config set provider copilot     # Use GitHub Copilot
+```
+
+Auto-prompts on first AI command if not set. Both CLIs must be in your PATH and authenticated.
+
+<details>
+<summary>Provider differences</summary>
+
+| Feature                     | Claude Code                          | GitHub Copilot                                                       |
+| --------------------------- | ------------------------------------ | -------------------------------------------------------------------- |
+| Status                      | GA                                   | Public preview                                                       |
+| Headless execution          | `-p --output-format json`            | `-p --output-format json --autopilot --no-ask-user`                  |
+| Session IDs                 | In JSON output (`session_id`)        | In JSON output (`session_id`), `--share` file as fallback            |
+| Session resume (`--resume`) | Full support                         | Full support                                                         |
+| Per-tool permissions        | Settings files + `--permission-mode` | `--allow-all-tools` (all-or-nothing by default)                      |
+| Fine-grained tool control   | `allow`/`deny` in settings files     | `--allow-tool`, `--deny-tool` flags (not yet used)                   |
+| Rate limit detection        | Validated patterns                   | Borrowed from Claude — not yet validated against real Copilot errors |
+
+</details>
 
 ---
 
-## CLI Overview
+## Data Directory
+
+All data lives in `~/.ralphctl/` by default. Override with:
+
+```bash
+export RALPHCTL_ROOT="/path/to/custom/data-dir"
+```
+
+---
+
+<details>
+<summary><strong>CLI Command Reference</strong></summary>
 
 ### Getting Started
 
@@ -135,7 +185,7 @@ Or just run `ralphctl` with no arguments for an interactive menu that walks you
 | ------------------------ | --------------------------------- |
 | `ralphctl sprint start`  | Execute tasks with AI             |
 | `ralphctl sprint health` | Diagnose blockers and stale tasks |
-| `ralphctl dashboard`     | Sprint overview with progress bar |
+| `ralphctl status`        | Sprint overview with progress bar |
 | `ralphctl task list`     | List tasks in the current sprint  |
 | `ralphctl task next`     | Show the next unblocked task      |
 | `ralphctl sprint close`  | Close an active sprint            |
@@ -143,54 +193,22 @@ Or just run `ralphctl` with no arguments for an interactive menu that walks you
 
 Run `ralphctl <command> --help` for details on any command.
 
----
-
-## AI Provider Configuration
-
-RalphCTL supports **Claude Code** and **GitHub Copilot** as AI backends. Both use the same prompt templates and
-workflow.
-
-```bash
-ralphctl config set provider claude      # Use Claude Code
-ralphctl config set provider copilot     # Use GitHub Copilot
-```
-
-Auto-prompts on first AI command if not set. Both CLIs must be in your PATH and authenticated.
-
-### Provider Differences
-
-| Feature                     | Claude Code                          | GitHub Copilot                                                       |
-| --------------------------- | ------------------------------------ | -------------------------------------------------------------------- |
-| Status                      | GA                                   | Public preview                                                       |
-| Headless execution          | `-p --output-format json`            | `-p --output-format json --autopilot --no-ask-user`                  |
-| Session IDs                 | In JSON output (`session_id`)        | In JSON output (`session_id`), `--share` file as fallback            |
-| Session resume (`--resume`) | Full support                         | Full support                                                         |
-| Per-tool permissions        | Settings files + `--permission-mode` | `--allow-all-tools` (all-or-nothing by default)                      |
-| Fine-grained tool control   | `allow`/`deny` in settings files     | `--allow-tool`, `--deny-tool` flags (not yet used)                   |
-| Rate limit detection        | Validated patterns                   | Borrowed from Claude — not yet validated against real Copilot errors |
+</details>
 
 ---
 
 ## Documentation
 
-| Document                                                    | Description                                    |
-| ----------------------------------------------------------- | ---------------------------------------------- |
-| [REQUIREMENTS.md](./.claude/docs/REQUIREMENTS.md)           | Acceptance criteria and feature requirements   |
-| [ARCHITECTURE.md](./.claude/docs/ARCHITECTURE.md)           | Data models, file storage, and error reference |
-| [CLAUDE.md](./CLAUDE.md)                                    | Developer guide and Claude Code project config |
-| [CONTRIBUTING.md](./CONTRIBUTING.md)                        | How to contribute                              |
-| [CHANGELOG.md](./CHANGELOG.md)                              | Version history                                |
-| [Blog post](https://lukasgrigis.dev/blog/building-ralphctl) | Background and motivation                      |
-
----
-
-## Data Directory
+| Resource                                       | Description                                |
+| ---------------------------------------------- | ------------------------------------------ |
+| [Architecture](./.claude/docs/ARCHITECTURE.md) | Data models, file storage, error reference |
+| [Requirements](./.claude/docs/REQUIREMENTS.md) | Acceptance criteria and feature checklist  |
+| [Contributing](./CONTRIBUTING.md)              | Dev setup, code style, PR process          |
+| [Changelog](./CHANGELOG.md)                    | Version history                            |
 
-RalphCTL stores all data in `~/.ralphctl/` by default. Override with `RALPHCTL_ROOT`:
+**Blog posts:** [Building ralphctl](https://lukasgrigis.dev/blog/building-ralphctl) (backstory) | [From task CLI to agent harness](https://lukasgrigis.dev/blog/ralphctl-agent-harness/) (evaluator deep-dive)
 
-```bash
-export RALPHCTL_ROOT="/path/to/custom/data-dir"
-```
+**Further reading:** [Harness Engineering for Coding Agent Users](https://martinfowler.com/articles/harness-engineering.html) — Martin Fowler (April 2026) | [Harness Design for Long-Running Application Development](https://www.anthropic.com/engineering/harness-design-long-running-apps) — Anthropic Engineering
 
 ---
 

package/dist/{add-SEDQ3VK7.mjs → add-DWNLZQ7Q.mjs}

@@ -2,12 +2,12 @@
 import {
   addSingleTicketInteractive,
   ticketAddCommand
-} from "./chunk-XPDI4SYI.mjs";
+} from "./chunk-QYF7QIZJ.mjs";
 import "./chunk-7TG3EAQ2.mjs";
-import "./chunk-LG6B7QVO.mjs";
-import "./chunk-KPTPKLXY.mjs";
+import "./chunk-7TBO6GOT.mjs";
+import "./chunk-ITRZMBLJ.mjs";
 import "./chunk-OEUJDSHY.mjs";
-import "./chunk-ZDEVRTGY.mjs";
+import "./chunk-GLDPHKEW.mjs";
 import "./chunk-EDJX7TT6.mjs";
 import "./chunk-QBXHAXHI.mjs";
 export {

package/dist/{add-TGJTRHIF.mjs → add-K7LNOYQ4.mjs}

@@ -2,12 +2,12 @@
 import {
   addCheckScriptToRepository,
   projectAddCommand
-} from "./chunk-Q3VWJARJ.mjs";
+} from "./chunk-LAERLCL5.mjs";
 import "./chunk-7LZ6GOGN.mjs";
 import "./chunk-7TG3EAQ2.mjs";
-import "./chunk-LG6B7QVO.mjs";
+import "./chunk-7TBO6GOT.mjs";
 import "./chunk-OEUJDSHY.mjs";
-import "./chunk-ZDEVRTGY.mjs";
+import "./chunk-GLDPHKEW.mjs";
 import "./chunk-EDJX7TT6.mjs";
 import "./chunk-QBXHAXHI.mjs";
 export {

package/dist/{chunk-LG6B7QVO.mjs → chunk-7TBO6GOT.mjs}

@@ -7,7 +7,7 @@ import {
   readValidatedJson,
   validateProjectPath,
   writeValidatedJson
-} from "./chunk-ZDEVRTGY.mjs";
+} from "./chunk-GLDPHKEW.mjs";
 import {
   ParseError,
   ProjectExistsError,

package/dist/{chunk-ZDEVRTGY.mjs → chunk-GLDPHKEW.mjs}

@@ -53,13 +53,20 @@ function getTasksFilePath(sprintId) {
 function getProgressFilePath(sprintId) {
   return join(getSprintDir(sprintId), "progress.md");
 }
+function assertSafeSegment(segment, label) {
+  if (!segment || segment.includes("/") || segment.includes("\\") || segment.includes("..") || segment.includes("\0")) {
+    throw new Error(`Path traversal detected in ${label}: ${segment}`);
+  }
+}
 function getRefinementDir(sprintId, ticketId) {
+  assertSafeSegment(ticketId, "ticket ID");
   return join(getSprintDir(sprintId), "refinement", ticketId);
 }
 function getPlanningDir(sprintId) {
   return join(getSprintDir(sprintId), "planning");
 }
 function getIdeateDir(sprintId, ticketId) {
+  assertSafeSegment(ticketId, "ticket ID");
   return join(getSprintDir(sprintId), "ideation", ticketId);
 }
 function getSchemaPath(schemaName) {
@@ -233,6 +240,7 @@ var TaskSchema = z.object({
   name: z.string().min(1),
   description: z.string().optional(),
   steps: z.array(z.string()).default([]),
+  verificationCriteria: z.array(z.string()).default([]),
   status: TaskStatusSchema.default("todo"),
   order: z.number().int().positive(),
   ticketId: z.string().optional(),
@@ -257,6 +265,7 @@ var ImportTaskSchema = z.object({
   // Required
   description: z.string().optional(),
   steps: z.array(z.string()).optional(),
+  verificationCriteria: z.array(z.string()).optional(),
   ticketId: z.string().optional(),
   blockedBy: z.array(z.string()).optional(),
   projectPath: z.string().min(1)

package/dist/{chunk-KPTPKLXY.mjs → chunk-ITRZMBLJ.mjs}

@@ -21,7 +21,7 @@ import {
   readValidatedJson,
   removeDir,
   writeValidatedJson
-} from "./chunk-ZDEVRTGY.mjs";
+} from "./chunk-GLDPHKEW.mjs";
 import {
   LockError,
   NoCurrentSprintError,

package/dist/{chunk-Q3VWJARJ.mjs → chunk-LAERLCL5.mjs}

@@ -8,7 +8,7 @@ import {
 } from "./chunk-7TG3EAQ2.mjs";
 import {
   createProject
-} from "./chunk-LG6B7QVO.mjs";
+} from "./chunk-7TBO6GOT.mjs";
 import {
   ensureError,
   wrapAsync
@@ -16,7 +16,7 @@ import {
 import {
   expandTilde,
   validateProjectPath
-} from "./chunk-ZDEVRTGY.mjs";
+} from "./chunk-GLDPHKEW.mjs";
 import {
   IOError,
   ProjectExistsError

package/dist/{chunk-AXNZMHFQ.mjs → chunk-ORVGM6EV.mjs}

@@ -11,7 +11,7 @@ import {
   getPendingRequirements,
   groupTicketsByProject,
   listTickets
-} from "./chunk-XPDI4SYI.mjs";
+} from "./chunk-QYF7QIZJ.mjs";
 import {
   EXIT_ALL_BLOCKED,
   EXIT_ERROR,
@@ -23,7 +23,7 @@ import {
 import {
   getProject,
   listProjects
-} from "./chunk-LG6B7QVO.mjs";
+} from "./chunk-7TBO6GOT.mjs";
 import {
   activateSprint,
   assertSprintStatus,
@@ -40,7 +40,7 @@ import {
   setAiProvider,
   summarizeProgressForContext,
   withFileLock
-} from "./chunk-KPTPKLXY.mjs";
+} from "./chunk-ITRZMBLJ.mjs";
 import {
   ensureError,
   unwrapOrThrow,
@@ -61,7 +61,7 @@ import {
   getTasksFilePath,
   readValidatedJson,
   writeValidatedJson
-} from "./chunk-ZDEVRTGY.mjs";
+} from "./chunk-GLDPHKEW.mjs";
 import {
   DependencyCycleError,
   IOError,
@@ -162,10 +162,13 @@ function buildEvaluatorPrompt(ctx) {
   const stepsSection = ctx.taskSteps.length > 0 ? `
 **Implementation Steps:**
 ${ctx.taskSteps.map((s) => `- ${s}`).join("\n")}` : "";
+  const criteriaSection = ctx.verificationCriteria.length > 0 ? `
+**Verification Criteria:**
+${ctx.verificationCriteria.map((c) => `- ${c}`).join("\n")}` : "";
   const checkSection = ctx.checkScriptSection ? `
 
 ${ctx.checkScriptSection}` : "";
-  return template.replaceAll("{{TASK_NAME}}", ctx.taskName).replace("{{TASK_DESCRIPTION_SECTION}}", descriptionSection).replace("{{TASK_STEPS_SECTION}}", stepsSection).replace("{{PROJECT_PATH}}", ctx.projectPath).replace("{{CHECK_SCRIPT_SECTION}}", checkSection);
+  return template.replaceAll("{{TASK_NAME}}", ctx.taskName).replace("{{TASK_DESCRIPTION_SECTION}}", descriptionSection).replace("{{TASK_STEPS_SECTION}}", stepsSection).replace("{{VERIFICATION_CRITERIA_SECTION}}", criteriaSection).replace("{{PROJECT_PATH}}", ctx.projectPath).replace("{{CHECK_SCRIPT_SECTION}}", checkSection);
 }
 
 // src/utils/requirements-export.ts
@@ -1087,6 +1090,7 @@ async function addTask(input3, sprintId) {
       name: input3.name,
       description: input3.description,
       steps: input3.steps ?? [],
+      verificationCriteria: input3.verificationCriteria ?? [],
       status: "todo",
       order: maxOrder + 1,
       ticketId: input3.ticketId,
@@ -1320,6 +1324,7 @@ function validateImportTasks(importTasks2, existingTasks, ticketIds) {
       name: t.name,
       description: void 0,
       steps: [],
+      verificationCriteria: [],
       status: "todo",
       order: existingTasks.length + i + 1,
       ticketId: void 0,
@@ -1355,7 +1360,7 @@ async function selectProject(message = "Select project:") {
       default: true
     });
     if (create) {
-      const { projectAddCommand } = await import("./add-TGJTRHIF.mjs");
+      const { projectAddCommand } = await import("./add-K7LNOYQ4.mjs");
       await projectAddCommand({ interactive: true });
       const updated = await listProjects();
       if (updated.length === 0) return null;
@@ -1428,7 +1433,7 @@ async function selectSprint(message = "Select sprint:", filter) {
       default: true
     });
     if (create) {
-      const { sprintCreateCommand } = await import("./create-DJHCP7LN.mjs");
+      const { sprintCreateCommand } = await import("./create-5MILNF7E.mjs");
       await sprintCreateCommand({ interactive: true });
       const updated = await listSprints();
       const refiltered = filter ? updated.filter((s) => filter.includes(s.status)) : updated;
@@ -1463,7 +1468,7 @@ async function selectTicket(message = "Select ticket:", filter) {
         default: true
       });
       if (create) {
-        const { ticketAddCommand } = await import("./add-SEDQ3VK7.mjs");
+        const { ticketAddCommand } = await import("./add-DWNLZQ7Q.mjs");
         await ticketAddCommand({ interactive: true });
         const updated = await listTickets();
         const refiltered = filter ? updated.filter(filter) : updated;
@@ -1658,6 +1663,7 @@ async function importTasksReplace(tasks, sprintId) {
       name: taskInput.name,
       description: taskInput.description,
       steps: taskInput.steps ?? [],
+      verificationCriteria: taskInput.verificationCriteria ?? [],
       status: "todo",
       order: newTasks.length + 1,
       ticketId: taskInput.ticketId,
@@ -2321,6 +2327,16 @@ function formatTask(ctx) {
       lines.push(`${String(i + 1)}. ${step}`);
     });
   }
+  if (ctx.task.verificationCriteria.length > 0) {
+    lines.push("");
+    lines.push("## Verification Criteria");
+    lines.push("");
+    lines.push("The task is done when all of the following are true:");
+    lines.push("");
+    ctx.task.verificationCriteria.forEach((criterion) => {
+      lines.push(`- ${criterion}`);
+    });
+  }
   return lines.join("\n");
 }
 function buildFullTaskContext(ctx, progressSummary, gitHistory, checkScript, checkStatus) {
@@ -2472,30 +2488,53 @@ function getEvaluatorModel(generatorModel, provider) {
   if (modelLower.includes("sonnet")) return "claude-haiku-4-5";
   return "claude-haiku-4-5";
 }
+var DIMENSION_NAMES = ["correctness", "completeness", "safety", "consistency"];
+var DIMENSION_PATTERNS = {
+  correctness: /\*\*correctness\*\*\s*:\s*(PASS|FAIL)\s*(?:—|-)\s*(.+)/i,
+  completeness: /\*\*completeness\*\*\s*:\s*(PASS|FAIL)\s*(?:—|-)\s*(.+)/i,
+  safety: /\*\*safety\*\*\s*:\s*(PASS|FAIL)\s*(?:—|-)\s*(.+)/i,
+  consistency: /\*\*consistency\*\*\s*:\s*(PASS|FAIL)\s*(?:—|-)\s*(.+)/i
+};
+function parseDimensionScores(output) {
+  const scores = [];
+  for (const dim of DIMENSION_NAMES) {
+    const match = DIMENSION_PATTERNS[dim].exec(output);
+    if (match?.[1] && match[2]) {
+      scores.push({
+        dimension: dim,
+        passed: match[1].toUpperCase() === "PASS",
+        finding: match[2].trim()
+      });
+    }
+  }
+  return scores;
+}
 function parseEvaluationResult(output) {
+  const dimensions = parseDimensionScores(output);
   if (output.includes("<evaluation-passed>")) {
-    return { passed: true, output };
+    return { passed: true, output, dimensions };
   }
   const failedMatch = /<evaluation-failed>([\s\S]*?)<\/evaluation-failed>/.exec(output);
   if (failedMatch) {
-    return { passed: false, output: failedMatch[1]?.trim() ?? output };
+    return { passed: false, output: failedMatch[1]?.trim() ?? output, dimensions };
   }
-  return { passed: false, output };
+  return { passed: false, output, dimensions };
 }
 function buildEvaluatorContext(task, checkScript) {
-  const checkScriptSection = checkScript ? `## Check Script
+  const checkScriptSection = checkScript ? `## Check Script (Computational Gate)
 
-You can run the following check script to verify the changes:
+Run this check script as the **first step** of your review \u2014 it is the same gate the harness uses post-task:
 
 \`\`\`
 ${checkScript}
 \`\`\`
 
-Run it to gain additional insight into whether the implementation is correct.` : null;
+If this script fails, the implementation fails regardless of code quality. Record the full output.` : null;
   return {
     taskName: task.name,
     taskDescription: task.description ?? "",
     taskSteps: task.steps,
+    verificationCriteria: task.verificationCriteria,
     projectPath: task.projectPath,
     checkScriptSection
   };
@@ -2520,6 +2559,7 @@ async function runEvaluation(task, generatorModel, checkScript, sprintId, provid
 }
 
 // src/ai/executor.ts
+var DEFAULT_MAX_TURNS = 200;
 function buildProviderArgs(options, provider) {
   if (provider.name !== "claude") {
     if (options.maxBudgetUsd != null) {
@@ -2528,6 +2568,9 @@ function buildProviderArgs(options, provider) {
     if (options.fallbackModel) {
       console.log(warning(`--fallback-model is only supported with the Claude provider \u2014 ignored`));
     }
+    if (options.maxTurns != null) {
+      console.log(warning(`--max-turns is only supported with the Claude provider \u2014 ignored`));
+    }
     return [];
   }
   const args = [];
@@ -2537,6 +2580,7 @@ function buildProviderArgs(options, provider) {
   if (options.fallbackModel) {
     args.push("--fallback-model", options.fallbackModel);
   }
+  args.push("--max-turns", String(options.maxTurns ?? DEFAULT_MAX_TURNS));
   return args;
 }
 async function executeTask(ctx, options, sprintId, resumeSessionId, provider, checkStatus) {
@@ -2672,6 +2716,8 @@ async function runEvaluationLoop(params) {
   const evalCheckScript = getEffectiveCheckScript(project, task.projectPath);
   const sprintDir = getSprintDir(sprintId);
   let evalResult = await runEvaluation(task, result.model, evalCheckScript, sprintId, provider);
+  let currentSessionId = result.sessionId;
+  let currentModel = result.model;
   for (let i = 0; i < evalIterations && !evalResult.passed; i++) {
     console.log(warning(`Evaluation failed for ${task.name} (iteration ${String(i + 1)}/${String(evalIterations)})`));
     console.log(muted(evalResult.output.slice(0, 500)));
@@ -2680,12 +2726,16 @@ async function runEvaluationLoop(params) {
       {
         cwd: task.projectPath,
         args: ["--add-dir", sprintDir, ...buildProviderArgs(options, provider)],
-        prompt: `The evaluator found issues with your work:
+        prompt: `The evaluator found issues with your implementation:
 
 ${evalResult.output}
 
-Fix these issues, then verify${options.noCommit ? "" : ", commit your fix,"} and signal completion.`,
-        resumeSessionId: result.sessionId ?? void 0,
+Review the critique carefully. Fix each identified issue in the code, then:
+1. Re-run verification commands to confirm the fix
+${options.noCommit ? "" : "2. Commit the fix with a descriptive message\n"}${options.noCommit ? "2" : "3"}. Signal completion with <task-verified> and <task-complete>
+
+If the critique is about something outside your task scope, fix only what is within scope and signal completion.`,
+        resumeSessionId: currentSessionId ?? void 0,
         env: provider.getSpawnEnv()
       },
       {
@@ -2699,6 +2749,8 @@ Fix these issues, then verify${options.noCommit ? "" : ", commit your fix,"} and
       provider
     );
     resumeSpinner?.succeed(`Fix attempt completed: ${task.name}`);
+    if (resumeResult.sessionId) currentSessionId = resumeResult.sessionId;
+    if (resumeResult.model) currentModel = resumeResult.model;
     const fixResult = parseExecutionResult(resumeResult.stdout);
     if (!fixResult.success) {
       console.log(warning(`Generator could not fix issues after feedback: ${task.name}`));
@@ -2712,7 +2764,7 @@ Fix these issues, then verify${options.noCommit ? "" : ", commit your fix,"} and
         break;
       }
     }
-    evalResult = await runEvaluation(task, resumeResult.model ?? result.model, evalCheckScript, sprintId, provider);
+    evalResult = await runEvaluation(task, currentModel, evalCheckScript, sprintId, provider);
   }
   await updateTask(
     task.id,
@@ -3797,6 +3849,16 @@ function parseArgs3(args) {
         throw new Error("Invalid model name \u2014 must be 1-100 alphanumeric characters, dots, hyphens, or underscores");
       }
       options.fallbackModel = modelStr;
+    } else if (arg === "--max-turns") {
+      const turnsStr = args[++i];
+      if (!turnsStr) {
+        throw new Error("--max-turns requires a number");
+      }
+      const turns = parseInt(turnsStr, 10);
+      if (isNaN(turns) || turns <= 0) {
+        throw new Error("--max-turns must be a positive integer");
+      }
+      options.maxTurns = turns;
     } else if (arg === "--no-evaluate") {
       options.noEvaluate = true;
     } else if (!arg?.startsWith("-")) {