@popoverai/dotrequirements 0.26.1 → 0.27.0

package/README.md

@@ -258,20 +258,12 @@ Then, in Claude Code, run `/codebase-to-spec` (or just ask — e.g. "spec the au
 
 ### `dotreq ai-setup`
 
-Configure the MCP server for AI assistants (Claude Code, Cursor, etc.).
+Install the requirements-driven workflow guidance for AI assistants (Claude Code, Cursor, Codex, Antigravity, GitHub Copilot) — and clean up any MCP configuration an earlier version wrote (the local MCP server is retired; agents use the CLI verbs directly).
 
 ```bash
 dotreq ai-setup
 ```
 
-### `dotreq mcp`
-
-Start the MCP server manually (typically not needed—AI assistants start it automatically after `ai-setup`).
-
-```bash
-dotreq mcp
-```
-
 ### `dotreq harness prepare`
 
 Parse requirements and build a lookup cache for multi-language test tracking. Run before tests in non-JavaScript projects. JavaScript projects don't need this — the test harness calls `prepare()` automatically.
@@ -332,10 +324,11 @@ dotreq style-check .requirements/auth.requirements.md --keys AUTH-LOGIN-1,AUTH-L
 
 | Option | Description |
 |--------|-------------|
+| `--source <local\|cloud>` | Where the review runs. Default `local` emits the style guide plus your content for your AI agent to judge — no account needed. `cloud` runs a hosted review (requires `dotreq link`). |
 | `--keys <list>` | Comma-separated requirement keys to limit the review (requirements files only) |
-| `--model <name>` | Override the AI model used for the review |
+| `--model <name>` | Override the AI model used for the review (`--source cloud` only) |
 
-When `.requirements/STYLE.md` exists in your project, its contents are included in the AI's prompt so project-specific style preferences influence the feedback. Requires cloud authentication.
+When `.requirements/STYLE.md` exists in your project, its contents are included so project-specific style preferences influence the feedback.
 
 ### `dotreq review-test`
 
@@ -345,7 +338,7 @@ AI-powered semantic review of a test file against its referenced requirements. V
 dotreq review-test src/auth.test.ts
 ```
 
-Catches tests that reference a requirement but don't actually validate what the requirement specifies — especially valuable when AI assistants write tests. Requires cloud authentication.
+Catches tests that reference a requirement but don't actually validate what the requirement specifies — especially valuable when AI assistants write tests. By default this emits the referenced requirements plus your test content for your AI agent to judge locally — no account needed. Add `--source cloud` for a hosted review (requires `dotreq link`).
 
 ### `dotreq create-requirement-document`
 
@@ -370,7 +363,7 @@ The following features work fully offline—no account required:
 - Reference requirements in tests (`requirement()`)
 - Local coverage reporting (`dotreq report`, default `--source local`)
 - Generate a requirements template (`dotreq create-requirement-document`)
-- MCP tools that mirror the offline CLI verbs
+- Local style-review materials for AI agents (`dotreq style-check`, `dotreq review-test`)
 
 The following features require a dot•requirements cloud account:
 
@@ -540,7 +533,7 @@ With cloud credentials configured, coverage is automatically reported to dot•r
 
 - Historical tracking of when requirements were last tested
 - Branch-based coverage (tracks `main`, feature branches, etc.)
-- Query coverage via the MCP server
+- Query cloud coverage from the terminal (`dotreq report --source cloud`)
 
 Run `dotreq link` to connect your project to the cloud and enable coverage reporting.
 
@@ -613,48 +606,19 @@ See [MARKDOWN_SCHEMA.md](https://github.com/PopoverAI/dotrequirements/blob/main/
 
 ---
 
-## MCP Server
-
-AI assistants can read your requirements in context, draft new ones, and verify tests actually validate what they claim to. The MCP server makes this possible through a standard protocol that works with Claude Code, Cursor, and other AI coding assistants.
-
-### Setup
-
-```bash
-dotreq ai-setup
-```
-
-Follow the prompts to configure for your AI assistant (Claude Code, Cursor, etc.).
+## AI Assistant Integration
 
-### Available Tools
+AI coding assistants use the CLI verbs directly — no MCP server to install or configure. `dotreq ai-setup` writes the requirements-driven workflow guidance into your assistant's context file (CLAUDE.md / AGENTS.md), which tells the agent when to explore, validate, style-review, and push. For chat apps (Claude, ChatGPT), use the dot•requirements remote connector instead.
 
-**Exploration:**
-- `search_requirements` — Search by text or regex
-- `get_requirement` — Get a requirement with its children and coverage
-- `list_requirements` — List all requirements in the project (set `untested: true` to filter to coverage gaps)
-- `get_requirements_by_test` — Get requirements referenced by a test file
-- `get_tests_by_requirement` — Get tests that reference a requirement
-
-**Authoring:**
-- `create_requirement_document` — Get a Markdown template with format examples
-- `validate_requirements` — Validate file schema (works offline)
-- `style_check` — AI-powered style feedback on requirements (supports optional `requirementKeys` filter)
-- `review_test` — Comprehensive test review (style + semantic correctness)
-
-**Coverage:**
-- `report_coverage` — Coverage from local cache or cloud, with optional requirement / branch / since filters
-
-**Cloud:**
-- `push_requirements` — Push to dot•requirements cloud
-
-**Diagnostic:**
-- `debug_mcp_environment` — Debug MCP server configuration
+> The local MCP server shipped by earlier versions is retired. `dotreq ai-setup` removes stale MCP registrations; `dotreq mcp` now exits with a pointer to this migration.
 
 ### CI/CD Mode
 
-For CI/CD environments (GitHub Actions, etc.), use the `--auth-from-env` flag to read credentials from environment variables instead of `project-settings.json`:
+For CI/CD environments (GitHub Actions, etc.), use the global `--auth-from-env` flag so cloud commands read credentials from environment variables instead of `project-settings.json`:
 
 ```bash
-node packages/cli/dist/mcp/index.js --auth-from-env
+dotreq --auth-from-env push
+dotreq --auth-from-env style-check .requirements/auth.requirements.md --source cloud
 ```
 
 Required environment variables:
@@ -663,23 +627,6 @@ Required environment variables:
 
 **Important:** Without `--auth-from-env`, these environment variables are ignored. This explicit opt-in prevents credential conflicts between local and CI environments.
 
-Example MCP config for GitHub Actions:
-
-```json
-{
-  "mcpServers": {
-    "dotrequirements": {
-      "command": "node",
-      "args": ["packages/cli/dist/mcp/index.js", "--auth-from-env"],
-      "env": {
-        "DOTREQ_PROJECT_ID": "${DOTREQ_PROJECT_ID}",
-        "DOTREQ_PROJECT_SECRET": "${DOTREQ_PROJECT_SECRET}"
-      }
-    }
-  }
-}
-```
-
 See the [CI/CD Integration docs](https://docs.dotrequirements.io/tools/ai/ci-cd) for complete examples.
 
 ---
@@ -698,9 +645,6 @@ import { parseRequirementsFile, validateRequirementsFile } from '@popoverai/dotr
 
 // Browser-compatible schema (no Node.js dependencies)
 import { parseRequirementBlock } from '@popoverai/dotrequirements/schema/browser';
-
-// MCP server
-import '@popoverai/dotrequirements/mcp';
 ```
 
 ---

package/dist/cli.js

@@ -24,6 +24,7 @@ import { styleCheckCommand } from "./commands/style-check.js";
 import { testsForCommand } from "./commands/tests-for.js";
 import { validateCommand } from "./commands/validate.js";
 import { loadEnvFile } from "./utils/env.js";
+import { setAuthFromEnv } from "./utils/project-settings.js";
 // Read version from package.json
 const __filename = fileURLToPath(import.meta.url);
 const __dirname = dirname(__filename);
@@ -54,8 +55,17 @@ function wrapCommand(fn) {
 const program = new Command();
 program
     .name("dotrequirements")
-    .description("Requirements tracking CLI with test harness and MCP server")
-    .version(VERSION);
+    .description("Requirements tracking CLI with test harness")
+    .version(VERSION)
+    // AUTHZ-6: explicit CI/CD credential injection — with this flag, cloud
+    // commands read DOTREQ_PROJECT_ID / DOTREQ_PROJECT_SECRET instead of
+    // project-settings.json; without it those variables are ignored.
+    .option("--auth-from-env", "Read cloud credentials from DOTREQ_PROJECT_ID and DOTREQ_PROJECT_SECRET (CI/CD)")
+    .hook("preAction", (thisCommand) => {
+    if (thisCommand.opts().authFromEnv) {
+        setAuthFromEnv(true);
+    }
+});
 program
     .command("init")
     .description("Initialize a new dotrequirements project")
@@ -126,11 +136,11 @@ program
     .action(wrapCommand(reportCommand));
 program
     .command("mcp")
-    .description("Start the MCP (Model Context Protocol) server for AI assistant integration")
+    .description("(retired) The local MCP server was removed — run `dotreq ai-setup` to update your configuration")
     .action(wrapCommand(mcpCommand));
 program
     .command("ai-setup")
-    .description("Configure MCP server for your AI assistant (Claude Code, Claude Desktop, etc.)")
+    .description("Set up your coding assistant to use the dotreq workflow (Claude Code, Cursor, Codex, Antigravity, GitHub Copilot)")
     .option("-a, --assistant <id>", "Configure for this assistant without prompting (e.g. claude-code)")
     .action(wrapCommand(aiSetupCommand));
 program
@@ -157,16 +167,18 @@ program
     .action(wrapCommand(testsForCommand));
 program
     .command("style-check <file>")
-    .description("AI style review of a requirements or test file (requires cloud auth)")
+    .description("Style review of a requirements or test file — emits judgment-ready review materials for the calling agent by default; --source cloud sends it for hosted AI review")
     .option("--keys <keys>", "Comma-separated requirement keys to limit the review (requirements files only)", (value) => value
     .split(",")
     .map((k) => k.trim())
     .filter(Boolean))
-    .option("--model <name>", "Override the AI model used for the review")
+    .option("--source <source>", 'Where the judgment runs: "local" (default) emits review materials for the caller; "cloud" uses the hosted review endpoint (requires cloud auth)')
+    .option("--model <name>", "Override the AI model used for the review (cloud mode only)")
     .action(wrapCommand(styleCheckCommand));
 program
     .command("review-test <test-file>")
-    .description("AI semantic review of a test file against its referenced requirements (requires cloud auth)")
+    .description("Semantic review of a test file against its referenced requirements — emits judgment-ready review materials for the calling agent by default; --source cloud sends it for hosted AI review")
+    .option("--source <source>", 'Where the judgment runs: "local" (default) emits review materials for the caller; "cloud" uses the hosted review endpoint (requires cloud auth)')
     .action(wrapCommand(reviewTestCommand));
 program
     .command("create-requirement-document [file-path]")

package/dist/codebase-to-spec/present.js

@@ -16,6 +16,7 @@
 import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs";
 import { dirname, join } from "node:path";
 import { parseRequirementsFile } from "../schema/parser.js";
+import { stripFrontmatterBlock } from "../schema/parser-core.js";
 import { generateRunMarker } from "../schema/run-marker.js";
 import { sanitizeAreaName } from "./area-name.js";
 import { promptOverwriteChoice } from "./interactive.js";
@@ -187,11 +188,9 @@ async function resolveOverwriteAction(policy, path) {
  */
 function mergeContent(existingPath, newContent) {
     const existing = readFileSync(existingPath, "utf-8").trimEnd();
-    // Strip frontmatter from newContent (it would conflict with existing frontmatter)
-    const fmMatch = newContent.match(/^---\n[\s\S]+?\n---\n+/);
-    const body = fmMatch
-        ? newContent.slice(fmMatch[0].length).trimStart()
-        : newContent;
+    // Strip frontmatter from newContent (it would conflict with existing
+    // frontmatter) via the shared CRLF-normalizing helper.
+    const body = stripFrontmatterBlock(newContent).trimStart();
     return [
         existing,
         "",

package/dist/codebase-to-spec/validate.js

@@ -87,8 +87,9 @@ function describeValidationError(err) {
         const parts = [err.message];
         let cursor = err.cause;
         while (cursor instanceof Error) {
-            // Don't repeat the same message twice if the wrapper just rethrew.
-            if (cursor.message && cursor.message !== parts[parts.length - 1]) {
+            // Don't repeat the same message twice if the wrapper just rethrew
+            // (or already embedded the inner message in its own).
+            if (cursor.message && !parts[parts.length - 1].includes(cursor.message)) {
                 parts.push(cursor.message);
             }
             cursor = cursor.cause;

package/dist/commands/acceptance-test.js

@@ -127,8 +127,10 @@ export async function acceptanceTestCommand(requirementKey, url, options) {
         else {
             displayResults(requirementTree, results);
         }
-        // 11. Exit with appropriate code
-        const failedCount = results.filter((r) => r.status !== "passed").length;
+        // 11. Exit with appropriate code — computed over the requirement tree so
+        // a requirement with a missing result counts as not passing, agreeing
+        // with the displayed counts (ACCEPTANCE-1.0/.1/.2)
+        const failedCount = requirementTree.filter((_, i) => results[i]?.status !== "passed").length;
         if (failedCount > 0) {
             process.exit(1);
         }

package/dist/commands/ai-setup.d.ts

@@ -1,13 +1,19 @@
 /**
  * Assistant identifiers accepted by the non-interactive --assistant flag
  * (AISETUP-1). "other" is interactive-only: it prompts for a custom path.
+ *
+ * Claude Desktop was dropped: it consumed only the retired local MCP server
+ * (no terminal, no context file), so setup has nothing to install for it. Its
+ * path forward is the remote MCP connector.
  */
-export declare const SUPPORTED_ASSISTANTS: readonly ["claude-code", "claude-desktop", "cursor", "antigravity", "codex", "github-copilot"];
+export declare const SUPPORTED_ASSISTANTS: readonly ["claude-code", "cursor", "antigravity", "codex", "github-copilot"];
 export interface AiSetupOptions {
     assistant?: string;
 }
 /**
- * AI setup command - configures the MCP server for AI assistants
+ * AI setup command — installs the requirements-driven workflow guidance into
+ * the assistant's context file, and removes any MCP configuration an earlier
+ * version wrote (the local MCP server is retired).
  *
  * AISETUP-1: with --assistant <id>, setup runs without any interactive prompt
  * (an AI assistant configuring itself knows which assistant it is).