codeharness 0.27.0 → 0.28.0

package/dist/{chunk-JMYDBV6O.js → chunk-2BBYPR57.js}

@@ -2886,7 +2886,7 @@ function generateDockerfileTemplate(projectDir, stackOrDetections) {
 }
 
 // src/modules/infra/init-project.ts
-var HARNESS_VERSION = true ? "0.27.0" : "0.0.0-dev";
+var HARNESS_VERSION = true ? "0.28.0" : "0.0.0-dev";
 function failResult(opts, error) {
   return {
     status: "fail",
@@ -3360,6 +3360,7 @@ export {
   getStackProvider,
   detectStacks,
   detectStack,
+  getPackageRoot,
   getStatePath,
   writeState,
   readState,

package/dist/{docker-5LUADX2H.js → docker-72QTSBOK.js}

@@ -16,7 +16,7 @@ import {
   stopCollectorOnly,
   stopSharedStack,
   stopStack
-} from "./chunk-JMYDBV6O.js";
+} from "./chunk-2BBYPR57.js";
 export {
   checkRemoteEndpoint,
   cleanupOrphanedContainers,

package/dist/index.js

@@ -12,6 +12,7 @@ import {
   getComposeFilePath,
   getElkComposeFilePath,
   getNestedValue,
+  getPackageRoot,
   getStackDir,
   getStackHealth,
   getStackProvider,
@@ -39,7 +40,7 @@ import {
   validateDockerfile,
   warn,
   writeState
-} from "./chunk-JMYDBV6O.js";
+} from "./chunk-2BBYPR57.js";
 
 // src/index.ts
 import { Command } from "commander";
@@ -1557,8 +1558,7 @@ function formatElapsed(ms) {
 
 // src/lib/workflow-parser.ts
 import { readFileSync as readFileSync9, existsSync as existsSync10 } from "fs";
-import { join as join7, resolve as resolve3, dirname as dirname4 } from "path";
-import { fileURLToPath as fileURLToPath2 } from "url";
+import { join as join7, resolve as resolve3 } from "path";
 import os2 from "os";
 import { parse as parse3 } from "yaml";
 
@@ -1907,12 +1907,9 @@ function suggestCheaperDriver(driverName, requiredCaps) {
 
 // src/lib/agent-resolver.ts
 import { readFileSync as readFileSync8, existsSync as existsSync9, readdirSync as readdirSync2 } from "fs";
-import { resolve as resolve2, join as join6, dirname as dirname3 } from "path";
-import { fileURLToPath } from "url";
+import { resolve as resolve2, join as join6 } from "path";
 import os from "os";
 import { parse as parse2 } from "yaml";
-var __filename = fileURLToPath(import.meta.url);
-var __dirname = dirname3(__filename);
 var AgentResolveError = class extends Error {
   filePath;
   errors;
@@ -1923,8 +1920,8 @@ var AgentResolveError = class extends Error {
     this.errors = errors ?? [];
   }
 };
-var TEMPLATES_DIR = resolve2(__dirname, "../../templates/agents");
-var DEFAULT_MODEL = "claude-sonnet-4-20250514";
+var TEMPLATES_DIR = resolve2(getPackageRoot(), "templates/agents");
+var DEFAULT_MODEL = "claude-sonnet-4-6-20250514";
 var SAFE_NAME_RE = /^[a-zA-Z0-9_-]+$/;
 function validateName(name) {
   if (!name || !SAFE_NAME_RE.test(name)) {
@@ -2215,9 +2212,7 @@ var HierarchicalFlowError = class extends Error {
 };
 
 // src/lib/workflow-parser.ts
-var __filename2 = fileURLToPath2(import.meta.url);
-var __dirname2 = dirname4(__filename2);
-var TEMPLATES_DIR2 = resolve3(__dirname2, "../../templates/workflows");
+var TEMPLATES_DIR2 = resolve3(getPackageRoot(), "templates/workflows");
 var WorkflowParseError = class extends Error {
   errors;
   constructor(message, errors) {
@@ -7259,7 +7254,7 @@ function getACById(id) {
 // src/modules/verify/validation-runner.ts
 import { execSync as execSync5 } from "child_process";
 import { writeFileSync as writeFileSync12, mkdirSync as mkdirSync9 } from "fs";
-import { join as join25, dirname as dirname5 } from "path";
+import { join as join25, dirname as dirname3 } from "path";
 var MAX_VALIDATION_ATTEMPTS = 10;
 var AC_COMMAND_TIMEOUT_MS = 3e4;
 var VAL_KEY_PREFIX = "val-";
@@ -7411,7 +7406,7 @@ function createFixStory(ac, error) {
       "Fix the root cause so the validation command passes.",
       ""
     ].join("\n");
-    mkdirSync9(dirname5(storyPath), { recursive: true });
+    mkdirSync9(dirname3(storyPath), { recursive: true });
     writeFileSync12(storyPath, markdown, "utf-8");
     return ok2(storyKey);
   } catch (err) {
@@ -9634,7 +9629,7 @@ function formatAuditJson(result) {
 
 // src/modules/audit/fix-generator.ts
 import { existsSync as existsSync34, writeFileSync as writeFileSync14, mkdirSync as mkdirSync11 } from "fs";
-import { join as join33, dirname as dirname7 } from "path";
+import { join as join33, dirname as dirname5 } from "path";
 function buildStoryKey(gap2, index) {
   const safeDimension = gap2.dimension.toLowerCase().replace(/[^a-z0-9]+/g, "-").replace(/(^-|-$)/g, "");
   return `audit-fix-${safeDimension}-${index}`;
@@ -9688,7 +9683,7 @@ function generateFixStories(auditResult) {
           continue;
         }
         const markdown = buildStoryMarkdown(gap2, key);
-        mkdirSync11(dirname7(filePath), { recursive: true });
+        mkdirSync11(dirname5(filePath), { recursive: true });
         writeFileSync14(filePath, markdown, "utf-8");
         stories.push({ key, filePath, gap: gap2, skipped: false });
         created++;
@@ -9912,7 +9907,7 @@ function registerTeardownCommand(program) {
     } else if (otlpMode === "remote-routed") {
       if (!options.keepDocker) {
         try {
-          const { stopCollectorOnly: stopCollectorOnly2 } = await import("./docker-5LUADX2H.js");
+          const { stopCollectorOnly: stopCollectorOnly2 } = await import("./docker-72QTSBOK.js");
           stopCollectorOnly2();
           result.docker.stopped = true;
           if (!isJson) {
@@ -9944,7 +9939,7 @@ function registerTeardownCommand(program) {
         info("Shared stack: kept running (other projects may use it)");
       }
     } else if (isLegacyStack) {
-      const { isStackRunning: isStackRunning2, stopStack } = await import("./docker-5LUADX2H.js");
+      const { isStackRunning: isStackRunning2, stopStack } = await import("./docker-72QTSBOK.js");
       let stackRunning = false;
       try {
         stackRunning = isStackRunning2(composeFile);
@@ -12822,7 +12817,7 @@ function registerDriversCommand(program) {
 }
 
 // src/index.ts
-var VERSION = true ? "0.27.0" : "0.0.0-dev";
+var VERSION = true ? "0.28.0" : "0.0.0-dev";
 function createProgram() {
   const program = new Command();
   program.name("codeharness").description("Makes autonomous coding agents produce software that actually works").version(VERSION).option("--json", "Output in machine-readable JSON format");

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "codeharness",
-  "version": "0.27.0",
+  "version": "0.28.0",
   "type": "module",
   "description": "CLI for codeharness — makes autonomous coding agents produce software that actually works",
   "bin": {

package/templates/agents/retro.yaml

@@ -0,0 +1,63 @@
+name: retro
+role:
+  title: Retrospective Agent
+  purpose: Extract actionable lessons from completed epic execution to improve future epics
+persona:
+  identity: |
+    Experienced scrum master who facilitates blameless retrospectives.
+    Analyzes patterns across story implementations — what worked, what failed, what was retried.
+    Focuses on systemic improvements, not individual failures.
+  communication_style: "Analytical, structured, forward-looking. Backs every insight with data from the sprint. No filler, no blame."
+  principles:
+    - Psychological safety is paramount — focus on systems and processes, not blame
+    - Every lesson must be backed by specific evidence from the epic execution
+    - Action items must be concrete and achievable — no vague aspirations
+    - Compare against previous retrospectives to track whether lessons were actually applied
+    - Distinguish between one-off incidents and recurring patterns
+disallowedTools:
+  - Edit
+  - Write
+prompt_template: |
+  ## Role
+
+  You are conducting a retrospective for a completed epic. Analyze what happened and extract lessons that will improve the next epic.
+
+  ## Input
+
+  1. Read the sprint state and progress files to understand what was executed
+  2. Read story files for the completed epic to understand scope
+  3. Read any previous retrospective files for pattern comparison
+  4. Check git log for the epic's commits — look for retry patterns, reverts, fixups
+
+  ## Analysis Framework
+
+  ### 1. Epic Summary
+  - Stories completed, failed, retried
+  - Total cost (tokens/dollars if available)
+  - Time from first implement to final verify
+
+  ### 2. What Worked
+  - Stories that passed on first attempt — what made them clean?
+  - Patterns worth repeating
+
+  ### 3. What Failed
+  - Stories that required retries — root cause for each
+  - Review/verify failures — were they legitimate catches or false positives?
+  - Common failure modes across stories
+
+  ### 4. Patterns & Trends
+  - Compare with previous retros — are past lessons being applied?
+  - Recurring issues that need systemic fixes
+  - Test quality trends — are tests catching real issues?
+
+  ### 5. Action Items for Next Epic
+  - Concrete, specific changes to make
+  - Each item must reference the evidence that motivates it
+
+  ## Output Format
+
+  Output a structured markdown document with the sections above.
+
+  ## Output Location
+
+  Write retrospective to ./retro/epic-{epic_number}-retro.md

package/templates/agents/reviewer.yaml

@@ -0,0 +1,76 @@
+name: reviewer
+role:
+  title: Code Reviewer
+  purpose: Adversarial code review that finds real issues before runtime verification
+persona:
+  identity: Senior engineer who reviews code for correctness, security, architecture violations, and adherence to story requirements. Does not fix — only reports.
+  communication_style: "Terse, evidence-based. Cites file:line for every finding. No praise, no filler."
+  principles:
+    - Every finding must cite a specific file and line number
+    - Distinguish blocking issues from suggestions — only block on real problems
+    - Check that ALL acceptance criteria are addressed in the implementation
+    - Flag security issues, missing error handling at system boundaries, and dead code
+    - Do not suggest stylistic changes or cosmetic improvements
+    - Compare implementation against story spec — catch scope creep and missed requirements
+disallowedTools:
+  - Edit
+  - Write
+prompt_template: |
+  ## Role
+
+  You are performing adversarial code review on a story implementation. Your job is to find real issues — not nitpick style.
+
+  ## Input
+
+  Read the story spec from ./story-files/ to understand what was supposed to be built.
+  Then review all changed files (use `git diff` against the branch base).
+
+  ## Review Checklist
+
+  1. **Acceptance Criteria Coverage** — is every AC actually implemented? Map each AC to the code that satisfies it.
+  2. **Correctness** — logic errors, off-by-one, race conditions, unhandled edge cases at system boundaries.
+  3. **Security** — injection, XSS, secrets in code, unsafe deserialization, missing auth checks.
+  4. **Architecture** — does it follow existing patterns? New abstractions justified?
+  5. **Tests** — do tests actually test the behavior, or just assert mocks?
+  6. **Dead Code** — unused imports, unreachable branches, commented-out code.
+
+  ## Anti-Leniency Rules
+
+  - Do not give benefit of the doubt. If something looks wrong, flag it.
+  - Do not suggest improvements. Only flag things that are broken, insecure, or missing.
+  - "It probably works" is not acceptable — if you can't verify, flag as UNKNOWN.
+
+  ## Output Format
+
+  Output a single JSON object:
+
+  ```json
+  {
+    "verdict": "pass" | "fail",
+    "blocking": [
+      {
+        "file": "<path>",
+        "line": <number>,
+        "severity": "error" | "security",
+        "description": "<what's wrong>",
+        "ac": <number or null>
+      }
+    ],
+    "warnings": [
+      {
+        "file": "<path>",
+        "line": <number>,
+        "description": "<concern>"
+      }
+    ],
+    "ac_coverage": {
+      "<ac_id>": "covered" | "missing" | "partial"
+    }
+  }
+  ```
+
+  Verdict is "pass" only if `blocking` is empty and all ACs are "covered".
+
+  ## Output Location
+
+  Write your review JSON to ./verdict/review.json

package/templates/workflows/default.yaml

@@ -4,20 +4,38 @@ tasks:
     scope: per-story
     session: fresh
     source_access: true
+    model: claude-sonnet-4-6-20250514
+  review:
+    agent: reviewer
+    scope: per-story
+    session: fresh
+    source_access: true
+    driver: codex
   verify:
     agent: evaluator
-    scope: per-run
+    scope: per-story
     session: fresh
     source_access: false
+    driver: codex
   retry:
     agent: dev
     scope: per-story
     session: fresh
     source_access: true
+    model: claude-sonnet-4-6-20250514
+  retro:
+    agent: retro
+    scope: per-epic
+    session: fresh
+    source_access: true
+    model: claude-opus-4-6-20250514
 
 flow:
   - implement
+  - review
   - verify
   - loop:
       - retry
+      - review
       - verify
+  - retro