clckernel 1.2.18 → 2.0.0

package/README.md

@@ -13,21 +13,25 @@
 
 AI coding agents (Cursor, Claude Code, Windsurf, Copilot) code fast, but without guardrails they create architectural decay, tautological tests, and secret leaks. 
 
-**CLC Kernel** is a polyglot developer harness that physically governs AI agents using **deterministic AST linters (< 5ms)**, **local SDD specs**, and an enforced **AI Unified Process (AIUP)**.
+**CLC Kernel** is a polyglot developer harness that governs AI agents using **deterministic pre-commit guards** (pattern-based checks, plus Python AST guards on Python stacks), **local SDD specs**, and an enforced **AI Unified Process (AIUP)**.
 
 ---
 
 ## 📋 Table of Contents
 - [🎯 Use Cases](#-use-cases)
 - [🏛️ The AI Unified Process (AIUP)](#️-the-ai-unified-process-aiup)
-- [🔍 Review-Driven Development (RDD) & the 4 Canonical Lenses](#-review-driven-development-rdd--the-4-canonical-lenses)
+- [🔍 Review-Driven Development (RDD) & the 4 Lenses](#-review-driven-development-rdd--the-4-lenses)
 - [⚡ Step-by-Step Guide (Full Lifecycle)](#-step-by-step-guide-full-lifecycle)
+- [🧭 What happens after install](#after-install)
 - [🎮 Universal Command Interface](#-universal-command-interface)
 - [🔄 Updating an Existing Project](#-updating-an-existing-project)
+- [⚙️ Configuration (`.clckernel.yaml`)](#configuration)
 - [🌐 Supported Stacks & Archetypes](#-supported-stacks--archetypes)
+- [⚠️ Known limitations](#known-limitations)
 - [🏛️ Architecture: Standalone vs. Gentle AI Symbiosis](#-architecture-standalone-vs-gentle-ai-symbiosis)
 - [🧩 Optional Ecosystem Companions](#-optional-ecosystem-companions)
 - [📖 Glossary of Key Terms](#-glossary-of-key-terms)
+- [🤝 Contributing & Community](#-contributing--community)
 - [📄 License](#-license)
 
 ---
@@ -37,10 +41,10 @@ AI coding agents (Cursor, Claude Code, Windsurf, Copilot) code fast, but without
 
 | Use Case | Problem It Solves | How CLC Kernel Solves It |
 |---|---|---|
-| **1. Coding Agent Governance**<br>*(Cursor, Claude Code, Copilot)* | Agents "vibe code", bypass tests, and violate layer boundaries in backend/frontend apps. | Enforces **Rule #0** (Research -> SDD -> TDD Red-to-Green -> AST Audit) and blocks commits via native AST guards. |
-| **2. Agent-OS & Domain Workflows**<br>*(Autonomous bots, marketing/sales OS)* | Agent runtimes execute actions without review, hallucinate metrics, or drift across IDE configs. | Provides **Dual-Layer Governance**, validates IDE mirror symlinks, checks Human-in-the-Loop (`check_hitl`), and enforces null-safe schemas (`check_domain_data`). |
-| **3. Complex Refactors & Migrations** | Agents modify files outside the agreed feature scope, causing silent regressions. | Locks work to a local `sdds/{change}/spec.md` and fails commits modifying untouched layers (`check_scope`). |
-| **4. Shift-Left CI/CD & Pre-commit** | Secret leaks, N+1 query loops, and broken database migrations reach remote branches. | Runs zero-dependency AST checks in your stack's native language in **< 5ms** before Git commit. |
+| **1. Coding Agent Governance**<br>*(Cursor, Claude Code, Copilot)* | Agents "vibe code", bypass tests, and violate layer boundaries in backend/frontend apps. | Enforces **Rule #0** (Research -> SDD -> TDD Red-to-Green -> Audit) and blocks commits via the pre-commit hook (`--no-verify` skips it; CI is the backstop). |
+| **2. Agent-OS & Domain Workflows**<br>*(Autonomous bots, marketing/sales OS)* | Agent runtimes execute actions without review, hallucinate metrics, or drift across IDE configs. | Provides **Dual-Layer Governance** and guards for any agent repository: explicit Human-in-the-Loop markers in skills and playbooks (`check_hitl`), SKILL.md frontmatter (`check_skills`), relative links (`check_links`), forgotten placeholders (`check_placeholders`), JSON data files that parse (`check_data_files`) and rule mirrors pointing to AGENTS.md (`check_symlinks`). Domain rules belong to the project: guards in `tools/guards/` run in the hook and the audit. |
+| **3. Complex Refactors & Migrations** | Agents modify files outside the agreed feature scope, causing silent regressions. | Anchors work to a local `sdds/{change}/spec.md`; scope is an agent-enforced protocol (not checked by a guard). |
+| **4. Shift-Left CI/CD & Pre-commit** | Secret leaks, N+1 query loops, and broken database migrations reach remote branches. | Runs zero-dependency checks in your stack's native language before Git commit: pattern-based guards everywhere, Python AST guards (layers, N+1 heuristic) on Python stacks. |
 
 ---
 
@@ -51,39 +55,51 @@ AI coding agents (Cursor, Claude Code, Windsurf, Copilot) code fast, but without
 ┌─────────────────────────────────────────────────────────────────────────────┐
 │                       AI UNIFIED PROCESS (AIUP)                             │
 ├─────────────┬─────────────┬─────────────┬──────────────────┬────────────────┤
-│   1. SDD    │   2. HIT    │   3. TDD    │     4. AST       │   5. MIRROR    │
+│   1. SDD    │   2. HIT    │   3. TDD    │    4. GUARDS     │   5. MIRROR    │
 │ Spec-Driven │ Human-in-the│ Red-to-Green│ Deterministic    │ Single Source  │
-│ Development │    Loop     │ Transitions │ Layer Linters    │ of Truth Sync  │
+│ Development │    Loop     │ Transitions │ Pre-commit Hook  │ of Truth Sync  │
 └─────────────┴─────────────┴─────────────┴──────────────────┴────────────────┘
 ```
 
 1. **SDD (Spec-Driven Development):** Agent must write a local, gitignored specification (`sdds/{feature}/spec.md`) before editing production files.
 2. **HIT (Human-in-the-Loop):** The engineer reviews and approves test scenarios and contracts before execution begins.
 3. **TDD (Red-to-Green):** Agent must prove failure with a failing regression test (Red Phase) before writing implementation code (Green Phase).
-4. **AST (Deterministic Safeguards):** Native stack parsers (Python `ast`, Go `ast`, RuboCop, TypeScript AST) inspect diffs and physically block invalid commits.
-5. **SSOT (Single Source of Truth):** `AGENTS.md` is forged once and mirrored via filesystem symlinks to `CLAUDE.md`, `GEMINI.md`, `.cursorrules`, and Copilot.
+4. **Guards (Deterministic Safeguards):** Pattern-based scripts (and Python `ast` parsers for Python stacks) inspect the staged diff; the pre-commit hook blocks non-compliant commits (`--no-verify` skips it; CI is the backstop).
+5. **SSOT (Single Source of Truth):** `AGENTS.md` is forged once and mirrored via filesystem symlinks to `CLAUDE.md`, `GEMINI.md`, `.cursorrules`, `.cursor/rules/clckernel_context.mdc`, `.github/copilot-instructions.md` and `.antigravity/rules.md` (the default `mirrors.list`).
 
 ---
 
 <a id="rdd-and-lenses"></a>
-## 🔍 Review-Driven Development (RDD) & the 4 Canonical Lenses
+## 🔍 Review-Driven Development (RDD) & the 4 Lenses
 
-CLC Kernel structures all verification and audit gates around **Review-Driven Development (RDD)**, categorizing every safeguard across the **4 Canonical Review Lenses**:
+CLC Kernel structures all verification and audit gates around **Review-Driven Development (RDD)**, with guard results grouped into **4 lenses** by a static name → lens table:
 
 ```
 ┌─────────────────────────────────────────────────────────────────────────────┐
-│                       THE 4 CANONICAL REVIEW LENSES                         │
+│                     RESULTS GROUPED INTO 4 REVIEW LENSES                    │
 ├─────────────────────┬───────────────────┬───────────────────┬───────────────┤
 │    🎯 CORRECTNESS   │    🛡️ SECURITY    │  🏛️ ARCHITECTURE  │ ⚡ PERFORMANCE│
-│ Functional bugs,    │ Secret leaks,     │ Layer violations, │ N+1 queries,  │
+│ Functional bugs,    │ Secret leaks,     │ Layer violations, │ N+1 heuristic,│
 │ scope drift, TDD,   │ permission bypass,│ domain boundaries,│ bundle size,  │
 │ API contracts       │ dangerous eval    │ circular deps     │ memory leaks  │
 └─────────────────────┴───────────────────┴───────────────────┴───────────────┘
 ```
 
-When you run `clckernel audit` (or `node tools/audit.js`), CLC Kernel generates:
-1. **Human-Readable Educational Report (`05-audit-report.md`):** Explains exactly what passed, failed, or was skipped grouped by lens.
-2. **Machine-Readable Audit Receipt (`audit-receipt.json`):** An immutable JSON receipt bound to the active `commitHash` certifying the verdict for automated review gates (such as Gentle AI).
+`clckernel audit` (or `node tools/audit.js` / `python3 tools/audit.py`) prints the 4-lens summary and writes
+nothing. Pass the SDD change it belongs to — `clckernel audit --change <name>` (or
+`node tools/audit.js --change <name>`) — and it also writes, in `sdds/<name>/` only:
+1. **Human-Readable Educational Report (`05-audit-report.md`):** Explains exactly what passed, failed, warned, or was skipped grouped by lens.
+2. **Machine-Readable Audit Receipt (`audit-receipt.json`):** An immutable JSON receipt bound to the active `commitHash` certifying the verdict (`PASSED`, `PASSED_WITH_WARNINGS` or `FAILED`) for automated review gates (such as Gentle AI).
+
+With both orchestrators installed (a mixed JS + Python stack), each writes its own pair so neither overwrites
+the other: `05-audit-report.js.md` / `audit-receipt.js.json` and `05-audit-report.py.md` / `audit-receipt.py.json`.
+
+Orchestrator CLI: `audit.(js|py) [targetDir] [--hook] [--change <name>] [--staged]`. `<name>` must be one path
+segment (letters, digits, `.`, `_`, `-`); `--staged` spells out the default snapshot (the staged changes, or
+the last commit when nothing is staged); `--hook` (used by the pre-commit hook) prints one line per guard and
+cannot be combined with `--change`; any other flag exits 2 with the usage. A failing guard whose level is
+`warning` — set by the guard (`level: 'warning'` in its result) or by `severities` in `.clckernel.yaml` — is
+printed as `WARN` and does not fail the audit or the commit; a guard that crashes always fails.
 
 ---
 
@@ -93,20 +109,21 @@ When you run `clckernel audit` (or `node tools/audit.js`), CLC Kernel generates:
 ```
 [1. Terminal]           [2. AI Chat]                 [3. TDD Loop]                 [4. Git Commit]
 npx clckernel   ───►   clckernel start   ───►   "Feature X with harness"   ───►   Pre-commit Hooks
-(Bootstrap)            (Interactive Setup)          (SDD -> HIT -> Tests)         (AST & Secret Guard)
+(Install)              (Interactive Setup)          (SDD -> HIT -> Tests)         (Guards & Secret Scan)
 ```
 
 ### 1️⃣ Step 1: Bootstrap the Repository (Terminal)
 ```bash
-npx clckernel start
+npx clckernel            # interactive: shows the detected stack and asks before writing
+npx clckernel . --yes    # non-interactive (add --dry-run to print the plan only)
 ```
-Auto-detects your framework, generates `AGENTS.md`, creates dynamic IDE symlinks (`CLAUDE.md`, `GEMINI.md`, `.cursorrules`), provisions pre-commit hooks, and creates native AST checks in `tools/`.
+Auto-detects your framework, generates `AGENTS.md`, creates the AI agent mirrors (symlinks to `AGENTS.md`: `CLAUDE.md`, `GEMINI.md`, `.cursorrules`, `.cursor/rules/clckernel_context.mdc`, `.github/copilot-instructions.md`, `.antigravity/rules.md`), writes the pre-commit hook, and installs native guard scripts in `tools/`. It ends with a **Next steps** block; see [What happens after install](#after-install).
 
 ### 2️⃣ Step 2: Initialize in Your AI Chat (Conversational CLI)
-Open Cursor, Claude Code, Gemini CLI, Windsurf, or Copilot and type:
+Open your AI agent (Claude Code, Gemini CLI, Cursor, Copilot, Antigravity, or any agent that reads `AGENTS.md`) and type:
 > `clckernel start`
 
-The agent scans your codebase silently, proposes an AIUP governance plan tailored to your framework, and generates `.clckernel.yaml`.
+`start` is an agent command (the CLI rejects it): the agent scans your codebase, proposes an AIUP governance plan tailored to your framework, installs the harness with `npx clckernel . --yes` if it is missing, and writes `.clckernel.yaml` only for settings you want to change (the file is optional). If you keep the defaults, it writes a `.clckernel.yaml` that holds only a comment, which ends the first-session onboarding.
 
 ### 3️⃣ Step 3: Develop with the AIUP Harness
 When requesting features or bugfixes, instruct your agent:
@@ -118,46 +135,104 @@ The agent autonomously executes the complete AIUP cycle:
 3. **HIT Review:** Requests your review and approval on proposed contracts.
 4. **TDD Red Phase:** Writes a failing regression test proving the requirement.
 5. **Clean Implementation (Green Phase):** Writes minimal code to pass tests while respecting layer isolation.
-6. **AST Audit Gate:** Runs `node tools/audit.js` and outputs the verification summary.
+6. **Audit Gate:** Runs `node tools/audit.js --change <name>` and outputs the verification summary (report in `sdds/<name>/`).
 
 ### 4️⃣ Step 4: Deterministic Guard Verification (Git Commit)
 ```bash
 git add .
 git commit -m "feat(auth): add user authentication endpoint"
 ```
-Pre-commit hooks automatically execute AST scanners (Clean Architecture, Secret Leaks, Scope Guard, Migration Idempotency). Any non-compliant commit is **physically blocked** before touching Git history.
+Pre-commit hooks automatically execute the installed guards (e.g. Clean Architecture, Secret Leaks, Migration Reversibility, depending on the stack). Any non-compliant commit is **blocked by the pre-commit hook** (`--no-verify` skips it; CI is the backstop).
+
+The secret scan (`scan_secrets`, on every stack) matches private keys, AWS, Stripe, OpenAI, Anthropic, Google, Slack and GitHub tokens, JWTs, quoted key/secret/password assignments, and secret-named values in committed `.env` files (`.env.example`, `.env.sample` and `.env.template` are skipped).
+
+The hook is written to `.githooks/pre-commit` (and `core.hooksPath` is set to `.githooks` in the repository's local git config), or to `.husky/pre-commit` when Husky already runs the repository's hooks. `core.hooksPath` is left unchanged when the directory is not a git repository root or another hook manager owns it; the Next steps block then prints the command that activates the hook. The hook does not list guards: it calls the audit orchestrators (`node tools/audit.js --hook`, `python3 tools/audit.py --hook`, each when installed) and then the stack's external tools (linters, test runners) when they are available. The orchestrators discover every guard — the kernel's and the project's own, either at the top of `tools/` (`check_*`, `scan_*`, `verify_*`, `*_guard`) or in `tools/guards/` — and run each exactly once, so the commit hook and `clckernel audit` always run the same set.
 
 ### 5️⃣ Step 5: Verify Repository Health Anytime (Doctor)
 ```bash
 npx clckernel doctor
 ```
-Audits 100% compliance across `AGENTS.md`, `sdds/`, `tools/`, active test runners, and Git hooks.
+Checks `AGENTS.md`, `sdds/`, `docs/Journal/`, `tools/`, the pre-commit hook, that each guard script loads (JS) or parses (Python), and the audit orchestrators' imports. It does not run the tests.
+
+---
+
+<a id="after-install"></a>
+## 🧭 What happens after install
+
+### In the terminal: Next steps
+A real install (not `--dry-run`) ends with a **Next steps** block built from what was installed. For a
+Next.js project in a git repository it reads:
+
+```text
+Next steps
+  1. Review and commit the harness: git add -A && git commit -m "chore: install clckernel"
+     The pre-commit hook (.githooks/pre-commit) runs on this commit.
+  2. Open your AI agent and ask it to run clckernel start to tailor the rules (.clckernel.yaml),
+     or just start working: it follows AGENTS.md.
+     AGENTS.md mirrors written for: Claude Code (CLAUDE.md), Gemini (GEMINI.md), Cursor, Copilot, Antigravity.
+  3. Run npx clckernel audit anytime to see every guard's result.
+  4. Project-specific checks go in tools/guards/ (the hook and the audit run them).
+  5. Later, preview kernel updates with npx clckernel update --dry-run.
+```
+
+- When the hook is not active (no git repository, a subdirectory of the repository, or `core.hooksPath`
+  owned by another hook manager), step 1 prints the activation command first, e.g.
+  `git init && git config core.hooksPath .githooks`, then the commit.
+- In a full-stack or monorepo install, step 1 adds that the hook runs the checks of every part from the
+  repository root.
+- Step 2 lists only the mirrors this run wrote or found up to date. A file of the same name that is not a
+  symlink (your own `CLAUDE.md`, for example) is kept, and that agent reads your file instead of `AGENTS.md`.
+  With no mirror written, step 2 says to use an agent that reads `AGENTS.md` itself.
+- `clckernel update` prints a shorter Next steps block only when something needs attention: a
+  `*.clckernel-new` copy to merge, or guards available with `--add-guards`.
+
+### In your AI agent: the first session
+Every AI harness reads the same law: Claude Code (`CLAUDE.md`), Gemini (`GEMINI.md`), Cursor
+(`.cursorrules`, `.cursor/rules/clckernel_context.mdc`), Copilot (`.github/copilot-instructions.md`) and
+Antigravity (`.antigravity/rules.md`) through the default mirrors; Windsurf through `.windsurfrules` when
+it is listed in `mirrors.list` at install time; agents such as Codex or OpenCode read `AGENTS.md` directly.
+
+Every generated `AGENTS.md` ends with a **First session after install** section. It applies while
+`tools/.clckernel-managed.json` exists and `.clckernel.yaml` does not:
+
+1. **Your request comes first.** If you ask for a task, the agent does it under the rules and mentions
+   onboarding once, in one sentence, at the end.
+2. **Otherwise it orients you briefly:** what the kernel installed (`tools/.clckernel-managed.json`) and what
+   the pre-commit hook runs on every commit.
+3. **It asks ONE question** and waits: tailor the rules now with `clckernel start`, or keep the defaults.
+   Keeping the defaults writes a `.clckernel.yaml` that holds only a comment, so the question is not asked
+   again.
+
+The `clckernel_cli` skill also triggers in that first session. Projects installed before this section
+existed do not get it on `update`: an existing `AGENTS.md` is never rewritten (update only adds the
+`clckernel update` row to its command table when missing). To get the section, copy it from a fresh
+install's `AGENTS.md`.
 
 ---
 
 <a id="universal-command-interface"></a>
 ## 🎮 Universal Command Interface (All LLMs & Agents)
 
-CLC Kernel provides a deterministic command grammar supported natively by **Claude Code, Cursor, Codeium/Windsurf, Copilot, Gemini, and Antigravity**:
+CLC Kernel defines a command grammar in `AGENTS.md`, which the agents read through their mirrors (**Claude Code, Gemini, Cursor, Copilot, Antigravity**; Windsurf when `.windsurfrules` is in `mirrors.list`) or directly (agents that read `AGENTS.md` themselves):
 
 | Command | Category | Mandatory Execution Protocol |
 |---|---|---|
 | `clckernel feature <name>` | Feature | Full AIUP cycle: Research -> Local Spec (`sdds/{name}/spec.md`) -> HIT review -> TDD Red Phase -> Green Implementation -> Audit. |
 | `clckernel fix <issue>` | Bug Fix | Root-cause analysis (no direct code edits) -> Failing regression test -> Surgical clean fix -> Audit gate. |
 | `clckernel test` | Verification | Runs repository test suite in the native stack runner and confirms clean pass. |
-| `clckernel audit` | Verification | Runs `node tools/audit.js` or `python tools/audit.py` and emits the Educational Summary. |
-| `clckernel doctor` | Diagnostics | Checks Git pre-commit hooks, IDE mirror symlinks, and AST engine integrity. |
+| `clckernel audit` | Verification | Runs `node tools/audit.js` and/or `python3 tools/audit.py` and emits the Educational Summary; with `--change <name>` it also writes the report and receipt to `sdds/<name>/`. |
+| `clckernel doctor` | Diagnostics | Runs `npx clckernel doctor [dir]`: checks AGENTS.md, sdds/, tools/, the pre-commit hook and that the guard scripts load. |
 | `clckernel commit <message>` | Version Control | Pre-flight audit, verifies diff vs SDD scope, and commits with Conventional Commits (zero AI attribution). |
-| `clckernel guard create <name>` | Safeguards | Scaffolds a new AST guard script in the stack's native language in `tools/guards/`. |
+| `clckernel guard create <name>` | Safeguards | Scaffolds a new guard script in the stack's native language in `tools/guards/`; the pre-commit hook and the audit run every project guard (`tools/guards/*.js`/`*.py`, and unmanaged `check_*`/`scan_*`/`verify_*` files at the top of `tools/`). |
 | `clckernel update` | Maintenance | Non-destructive update: refreshes core tools, skills & IDE symlinks without overwriting SDDs, custom guards, or `.clckernel.yaml`. |
-| `clckernel start` | Setup | Auto-detects stack, proposes AIUP lifecycle, and generates `.clckernel.yaml`. |
+| `clckernel start` | Setup | Agent command: auto-detects the stack, proposes the AIUP lifecycle, installs the harness (`npx clckernel . --yes`) if missing; `.clckernel.yaml` is optional and written only for non-default settings (keeping the defaults writes a comment-only file). |
 
 ---
 
 <a id="updating-an-existing-project"></a>
 ## 🔄 Updating an Existing Project (Non-Destructive)
 
-When new versions of CLC Kernel are released with new AST guards, upgraded playbooks, or improved IDE rules, update your project safely without losing your work:
+When new versions of CLC Kernel are released with new guards, upgraded playbooks, or improved IDE rules, update your project safely without losing your work:
 
 ```bash
 npx clckernel update
@@ -168,13 +243,87 @@ clckernel update
 ```
 
 ### 🔒 The Non-Destructive Update Guarantee:
-- **Refreshes & Upgrades:** Core orchestrators (`tools/audit.js`, `tools/audit.py`), built-in AST guards, Agent Skills (`.agents/skills/`), and IDE symlinks (`CLAUDE.md`, `GEMINI.md`, `.cursorrules`, Copilot, Antigravity).
+- **Refreshes & Upgrades:** Core orchestrators (`tools/audit.js`, `tools/audit.py`), the built-in guards your project has, Agent Skills (`.agents/skills/`), and the IDE symlinks your project has (`CLAUDE.md`, `GEMINI.md`, `.cursorrules`, Copilot, Antigravity).
 - **100% Preserved:**
   - `sdds/`: All local specifications, plans, and verification receipts remain intact.
-  - `tools/guards/`: All team-authored custom AST guards are preserved without modification.
-  - `.clckernel.yaml`: All custom thresholds and active safeguard selections are preserved.
+  - Project guards (`tools/guards/`, and guard files at the top of `tools/` the kernel does not own) are preserved without modification. A kernel guard the kernel no longer ships stays exactly where it is, leaves the registry and keeps running as a project guard; `update` reports it as a "retired kernel guard kept as project guard".
+  - `.clckernel.yaml`: never rewritten; update reads it (see [Configuration](#configuration)).
   - Application source code and domain models are never touched.
 
+### 🛡️ Files you edited are never overwritten
+`update` (and a reinstall with `--yes`) only overwrites a kernel file (`tools/*`, the pre-commit hook, the
+`clckernel_cli` skill) when it can prove the file is its own and intact: its sha256 matches the registry
+(`tools/.clckernel-managed.json`) or a version the kernel published. Anything else — a guard you hardened, a
+project file that happens to share a kernel file's name — is kept byte for byte, and the new version is
+written next to it as `<file>.clckernel-new` (gitignored) for you to review and merge. The summary lists what
+was written, unchanged, protected, moved, added to `.gitignore` and set in `git config`.
+
+Preview first with `--dry-run`: it prints the same summary as a plan and writes nothing (no files,
+`.gitignore` or `git config`):
+```bash
+npx clckernel update . --dry-run
+npx clckernel . --yes --dry-run
+```
+
+### ➕ Update never adds what you did not choose
+`update` refreshes what the project already has; it never adds guards or IDE mirrors on its own:
+- **Guards:** only the guards in the registry or present in `tools/` are refreshed and run by the hook. New
+  guards of your stack are listed as *available, not installed*; opt in with `--add-guards` (they stay
+  installed and updated from then on). An unknown name fails with exit 1 and writes nothing. Support files
+  (`_changes.js/.py`, the config readers `_config.js/.py`, `yaml.js`, `_yaml.py`, the audit orchestrators)
+  follow the installed guards.
+- **Mirrors:** a mirror you deleted (e.g. `.cursor/rules/clckernel_context.mdc`) is not recreated; it is
+  reported as *not present (skipped)*. A fresh install still creates them all.
+- **`.gitignore`:** an entry whose path git tracks (a mirror or the skill you commit) is never added, on
+  install or update; it is reported as *tracked, not ignored*.
+```bash
+npx clckernel update . --add-guards check_skills,check_placeholders
+npx clckernel update . --add-guards all
+```
+
+---
+
+<a id="configuration"></a>
+## ⚙️ Configuration (`.clckernel.yaml`)
+
+Optional. Without the file every default below applies (it is the behaviour of a project without
+config). Install, update and every guard read it from the project root; an invalid value stops them
+with `Invalid .clckernel.yaml: key "<key>" — <reason>` (nothing is written, the commit is blocked).
+
+| Key | Effect | Default | Read by |
+|---|---|---|---|
+| `rules` | Custom regex rules (`name`, `pattern`, `message`, optional `severity`, `scope`, `exclude`) run against the changed files | none | `check_custom` |
+| `exclude_paths` | Globs no diff-based guard evaluates (secret scan, TDD verifier, custom rules, …) | none | `_changes.js/.py` |
+| `mirrors.list` | AI IDE mirrors (symlinks to `AGENTS.md`) install creates and `check_symlinks` verifies | install: `CLAUDE.md`, `GEMINI.md`, `.cursorrules`, `.cursor/rules/clckernel_context.mdc`, `.github/copilot-instructions.md`, `.antigravity/rules.md`; `check_symlinks`: `CLAUDE.md`, `GEMINI.md`, `.cursorrules`, `.windsurfrules` | install/update, `check_symlinks` |
+| `mirrors.required` | `true`: a listed mirror that is missing fails `check_symlinks` | `false` | `check_symlinks` |
+| `hitl.markers` | Regexes (case-insensitive, one line, outside code fences) that count as a Human-in-the-Loop marker; they replace the defaults | a `HITL:` line or a `## Human checkpoint` / `## HITL` heading | `check_hitl` |
+| `hitl.paths` | Globs of the changed files `check_hitl` evaluates | skills (`SKILL.md`, flat `skills/*.md`) and `playbooks/**/*.md` | `check_hitl` |
+| `hitl.optOut` | `{ key, value }`: a file whose frontmatter has `key: value` is exempt | none | `check_hitl` |
+| `update.exclude` | Root-relative globs install and update never write, move or remove; listed as *excluded by config* | none | install/update |
+| `tdd.code` | Extra globs counted as production code (any language), e.g. Python tools tested by JS tests | none | `verify_tdd.js/.py` |
+| `tdd.ignore` | Globs never counted as production code | none | `verify_tdd.js/.py` |
+| `severities` | `{ <guard name>: warning \| error }`: a failing guard at `warning` is reported (`WARN`, verdict `PASSED_WITH_WARNINGS`) but does not block; `error` blocks even when the guard reports `level: 'warning'`. A crash always blocks | each guard's own level (`error` unless it reports `warning`) | `audit.js` / `audit.py` (audit and pre-commit hook) |
+
+```yaml
+exclude_paths: ["vendor/**"]
+mirrors:
+  list: [CLAUDE.md, .github/copilot-instructions.md]
+  required: true
+hitl:
+  optOut: { key: checkpoint, value: none }
+update:
+  exclude: [AGENTS.md]
+tdd:
+  code: ["tools/**/*.py"]
+severities:
+  check_seo: warning
+```
+
+Keys earlier releases accepted without any effect — `active_guards`, `phases`, `gate_mode`,
+`layers`, `scope` — are ignored with a deprecation warning; remove them. An unknown top-level key is
+ignored with a warning. The guards read it through `tools/_config.js` (+ `yaml.js`) and `tools/_config.py`
+(+ `_yaml.py`), installed next to `_changes.js/.py`, so a file created after install applies at once.
+
 ---
 
 <a id="supported-stacks"></a>
@@ -183,14 +332,81 @@ clckernel update
 | Ecosystem | Detection Signature | Test Runner | Native Safeguards |
 |---|---|---|---|
 | ⚛️ **Next.js** | `next` | Vitest / Jest | Clean Arch, UI Reuse, Semantic Tokens, RSC, A11y, Responsive, SEO |
-| ⚡ **FastAPI** | `fastapi` | Pytest | Pydantic V2, Alembic Idempotency, Clean Arch AST, DB Efficiency (N+1) |
-| 💎 **Rails** | `Gemfile` | RSpec | RuboCop AST, Brakeman Security, Migration Idempotency |
-| 🐹 **Golang** | `go.mod` | `go test` | `golangci-lint` AST, Domain/UseCase Layer Isolation |
-| 🦀 **Rust** | `Cargo.toml` | `cargo test` | `cargo clippy` AST, `cargo audit`, Strict Memory Safety |
-| 🐘 **Laravel** | `artisan` | Pest / PHPUnit | PHPStan AST, Eloquent N+1 Loop Detection, Migrations |
-| 🎸 **Django** | `manage.py` | `pytest-django` | Django ORM Idempotency, Ruff AST, Scope Guard, DB Efficiency (N+1) |
+| ⚡ **FastAPI** | `fastapi` | Pytest | Alembic Migration Reversibility, Clean Arch (Python AST), N+1 Heuristic (Python AST), Red→Green TDD Verifier (pytest) |
+| 💎 **Rails** | `Gemfile` | RSpec | RuboCop, Brakeman Security, Red→Green TDD Verifier (RSpec / Minitest) |
+| 🐹 **Golang** | `go.mod` | `go test` | `golangci-lint`, Domain/UseCase Layer Isolation, Red→Green TDD Verifier (`go test`) |
+| 🦀 **Rust** | `Cargo.toml` | `cargo test` | `cargo clippy`, `cargo audit`, Strict Memory Safety, Red→Green TDD Verifier (`cargo test`, integration tests) |
+| 🐘 **Laravel** | `artisan` | Pest / PHPUnit | PHPStan static analysis, Red→Green TDD Verifier (Pest / PHPUnit) |
+| 🎸 **Django** | `manage.py` | `pytest-django` | Ruff, Clean Arch (Python AST), N+1 Heuristic (Python AST), Red→Green TDD Verifier (pytest) |
 | 🚀 **Astro** | `astro.config` | Playwright | Tailwind v4 Tokens, Content Schema, A11y, Responsive, SEO |
-| 🤖 **Agent-OS** | `skills/`, `brand/`, `templates/` | Contract Tests | Dual-Layer AGENTS.md, HITL Gates (`check_hitl`), Symlinks (`check_symlinks`), Domain Data (`check_domain_data`) |
+| 🤖 **Agent-OS** | `skills/<name>/SKILL.md` (also `.agents/skills/`, `.claude/skills/`), `playbooks/` or `rules/` with Markdown; a code stack in the same repo wins | Contract Tests | Dual-Layer AGENTS.md, HITL markers (`check_hitl`), Skills (`check_skills`), Links (`check_links`), Placeholders (`check_placeholders`), JSON data (`check_data_files`), Mirrors (`check_symlinks`), Custom rules (`check_custom`); project guards in `tools/guards/` |
+
+### 🧱 Full-stack & monorepos
+
+A repository can hold several code parts. `clckernel` looks for a stack in the root, in the directories
+declared as workspaces (`package.json#workspaces`, array or `{ packages }`, and `pnpm-workspace.yaml`
+`packages:`) and in the conventional directories `frontend`, `backend`, `web`, `api`, `server`, `client`,
+`apps/*`, `services/*` and `packages/*` (at most two levels deep; `node_modules`, `.git`, `vendor`, `dist` and
+`build` are skipped). Workspace patterns are read as literal directories or `dir/*`; other glob forms are
+ignored.
+
+- **One part at the root:** installed exactly as a single-stack repository.
+- **Several parts** (e.g. Next.js at the root + `backend/` FastAPI, or `apps/web` + `apps/api`): the project
+  type is `fullstack` (a part at the root) or `monorepo` (only subdirectories). Two stacks of the same
+  ecosystem in one directory (e.g. FastAPI and Django) stay one part, the first adapter winning as before.
+- **Guards:** the union of the parts' guards is installed once in the root `tools/`, with one audit
+  orchestrator per guard language, and runs once from the root.
+- **External tools** (ruff, pytest, tsc, …) run inside their part's directory; a missing one prints
+  `clckernel: <tool> (<dir>) not available, skipped`.
+- **Red→Green TDD Verifier:** each changed file belongs to its nearest project root (Python:
+  `pyproject.toml`, `requirements.txt`, `setup.cfg`, `setup.py`, `manage.py`; JS: `package.json`; Go: `go.mod`;
+  Rust: `Cargo.toml`; Ruby: `Gemfile`; PHP: `composer.json`). Each group runs in its root (`PYTHONPATH`, runner
+  and `node_modules` of that root, or the repository root's `node_modules` when hoisted). Code changed in a part
+  without a changed test in that same part fails.
+- **AGENTS.md** lists every part (directory, stack, test runner) with each stack's rules; `clckernel update`
+  re-detects the parts.
+
+<a id="red-green-other-stacks"></a>
+#### Red→Green on Go, Rust, Ruby and PHP
+
+`tools/verify_tdd.js` (Node is present on every stack) runs the changed tests on the staged code (must pass,
+with at least one test executed) and on `HEAD` plus those tests (must fail), from trees materialized from git:
+
+| Language | Tests | Production code | Command |
+|---|---|---|---|
+| Go | `*_test.go` | other `.go` (not `vendor/`) | `go test -v -count=1 ./<pkg>…` (packages of the changed tests) |
+| Rust | `tests/*.rs`, `tests/<t>/main.rs` | `src/**/*.rs` | `cargo test --test <t>…` (temporary `CARGO_TARGET_DIR`) |
+| Ruby | `spec/**/*_spec.rb`, `test/**/*_test.rb` | `app/**/*.rb`, `lib/**/*.rb` | `bundle exec rspec` / `bundle exec ruby -Ilib -Itest` |
+| PHP | `tests/**/*Test.php` | `app/**/*.php`, `src/**/*.php` | `vendor/bin/pest`, else `vendor/bin/phpunit` (one file per run) |
+
+A missing `go`, `cargo`, `bundle`, locked `rspec-core` (for specs) or `vendor/bin/{pest,phpunit}` fails the
+check with an install hint. Limitations: Rust `#[cfg(test)]` unit tests live inside the code file and cannot be
+separated for the red run, so a `src/` change needs a changed integration test under `tests/` (or a
+`tdd.ignore` entry); Minitest runs through plain `ruby -Ilib -Itest`, not `rails test`; Go is verified per
+package; PHP assumes Composer 2.2+ bin proxies. Only committed or staged files exist in the trees (no `.env`).
+
+Not supported: Agent-OS as one part of a monorepo, nested git repositories or submodules, choosing parts by
+hand in the interactive prompt ("Choose another stack" installs a single stack), and Go/Cargo/Composer
+workspaces (their parts are found only through the conventional directories).
+
+---
+
+<a id="known-limitations"></a>
+## ⚠️ Known limitations
+
+- **Windows is not tested.** The pre-commit hook is a shell script (`#!/usr/bin/env bash`, or `sh` under
+  Husky): on Windows it needs a bash/sh such as Git Bash.
+- **Your own lint-staged / ESLint config may lint the harness files under `tools/`.** The kernel's files are
+  listed in `tools/.clckernel-managed.json`; exclude `tools/` (or those paths) in your linter config.
+- **Rust inline unit tests** (`#[cfg(test)]`) cannot be separated for the red run; see
+  [Red→Green on Go, Rust, Ruby and PHP](#red-green-other-stacks).
+- **Existing installs keep their `AGENTS.md`.** `update` never rewrites it, so the *First session after
+  install* section and other template changes reach only fresh installs; copy them by hand if you want them.
+- **Windsurf has no default mirror.** List `.windsurfrules` in `mirrors.list` before installing; on an
+  existing install `update` does not add mirrors, so create it yourself (`ln -s AGENTS.md .windsurfrules`).
+- **External linters and test runners run only when available.** The hook runs each one (tsc, ruff, pytest,
+  golangci-lint, cargo, …) when the project has it at commit time and otherwise prints
+  `clckernel: <tool> not available, skipped`; the guards in `tools/` always run.
 
 ---
 
@@ -210,17 +426,17 @@ flowchart TD
   subgraph CORE["🤖 CLC KERNEL (Core Governance Engine)"]
     SSOT["1. SSOT Law<br/>(AGENTS.md & IDE Mirrors)"]
     AIUP["2. AIUP Harness<br/>(SDD + HITL + TDD Red-to-Green)"]
-    AST["3. AST Guards &lt;5ms<br/>(Layer, Secret & Migration Rules)"]
+    GUARDS["3. Pre-commit Guards<br/>(Layer, Secret & Migration Rules)"]
     RECEIPT["4. 4-Lens Audit Engine<br/>(Emits audit-receipt.json)"]
-    SSOT --> AIUP --> AST --> RECEIPT
+    SSOT --> AIUP --> GUARDS --> RECEIPT
   end
 
   RECEIPT -->|audit-receipt.json| MODES{Deployment Mode}
 
   subgraph MODE1["🟢 MODE 1: STANDALONE (100% Local & Offline)"]
     M1_1["• Zero foreign dependencies"]
-    M1_2["• Native stack AST execution in &lt; 5ms"]
-    M1_3["• Git pre-commit hook physically blocks invalid commits"]
+    M1_2["• Guards run in the stack's native language"]
+    M1_3["• Git pre-commit hook blocks invalid commits (--no-verify skips it; CI is the backstop)"]
     M1_4["• Self-contained 05-audit-report.md"]
   end
 
@@ -254,17 +470,25 @@ CLC Kernel focuses strictly on governance and runs with **zero foreign dependenc
 |---|---|
 | **SDD** *(Spec-Driven Development)* | A specification-first methodology where detailed technical contracts, schemas, acceptance criteria, and test scenarios are authored in local, gitignored specs (`sdds/{change}/spec.md`) before modifying source code. |
 | **RDD** *(Review-Driven Development)* | An adversarial and structured code verification methodology where changes are audited across four orthogonal lenses (Correctness, Security, Architecture, Performance) to ensure systematic quality gates before merge or commit. |
-| **AIUP** *(AI Unified Process)* | The end-to-end governed lifecycle implemented by CLC Kernel: Research $\rightarrow$ Local SDD $\rightarrow$ Human-in-the-Loop (HITL) approval $\rightarrow$ TDD Red-to-Green $\rightarrow$ AST Guard Audit Gate $\rightarrow$ Conventional Commit. |
+| **AIUP** *(AI Unified Process)* | The end-to-end governed lifecycle implemented by CLC Kernel: Research $\rightarrow$ Local SDD $\rightarrow$ Human-in-the-Loop (HITL) approval $\rightarrow$ TDD Red-to-Green $\rightarrow$ Guard Audit Gate $\rightarrow$ Conventional Commit. |
 | **HIT / HITL** *(Human-in-the-Loop)* | A mandatory approval gate where a human engineer validates contracts, architecture decisions, and proposed tests before an autonomous agent executes production code edits. |
 | **TDD** *(Test-Driven Development)* | The engineering practice of writing a failing automated regression test (Red Phase) to reproduce an issue or specify a new feature before writing the minimal production code to make it pass (Green Phase). |
-| **AST Guard** *(Abstract Syntax Tree Safeguard)* | A lightweight, zero-dependency static analysis script in the stack's native language (Python `ast`, TypeScript Compiler API, Go `go/parser`, RuboCop, etc.) executing in under 5ms to physically block architectural violations before git commits. |
-| **SSOT** *(Single Source of Truth)* | The centralized governance principle where `AGENTS.md` acts as the single source of truth, mirrored automatically to all AI IDE config files (`CLAUDE.md`, `GEMINI.md`, `.cursorrules`, Copilot) via filesystem symlinks. |
+| **Guard** *(Safeguard script)* | A lightweight, zero-dependency check in the stack's native language, run by the pre-commit hook. Most guards are pattern-based (regex); on Python stacks the layer, N+1 and Celery/Redis/Postgres guards parse code with Python `ast`. The hook blocks violations before commit (`--no-verify` skips it; CI is the backstop). |
+| **SSOT** *(Single Source of Truth)* | The centralized governance principle where `AGENTS.md` acts as the single source of truth, mirrored to the AI agent rule files (`CLAUDE.md`, `GEMINI.md`, `.cursorrules`, `.cursor/rules/clckernel_context.mdc`, `.github/copilot-instructions.md`, `.antigravity/rules.md`; configurable with `mirrors.list`) via filesystem symlinks. |
 | **Audit Receipt** | An immutable, machine-readable JSON artifact (`audit-receipt.json`) produced by the audit engine containing the timestamp, commit hash, per-lens verdicts, and Gentle AI RDD compatibility metadata. |
 | **ADR** *(Architectural Decision Record)* | A standardized document in `docs/Journal/` capturing significant architectural decisions, context, evaluated alternatives, and consequences. |
-| **Agent-OS** | An operating system archetype for non-tech domain workflows (e.g. Marketing OS, Content Ops, Research OS) governed by dual-layer architecture, strict schema validation, and HITL checkpoints. |
+| **Agent-OS** | An operating system archetype for non-tech domain workflows (e.g. Marketing OS, Content Ops, Research OS) made of skills, playbooks, rules, prompts and templates, governed by a dual-layer AGENTS.md, explicit HITL checkpoints and document guards (skill frontmatter, links, placeholders). |
 
 ---
 
+## 🤝 Contributing & Community
+Contributions are welcome. Start with these documents:
+
+- [CONTRIBUTING.md](CONTRIBUTING.md) — setup, the pre-commit hook, tests and the pull request flow.
+- [CODE_OF_CONDUCT.md](CODE_OF_CONDUCT.md) — expected behavior and how to report a problem.
+- [SECURITY.md](SECURITY.md) — how to report a vulnerability privately.
+- [CHANGELOG.md](CHANGELOG.md) — notable changes per release.
+
 <a id="license"></a>
 ## 📄 License
-MIT © [CarlosLeonCode](https://github.com/carlosleoncode)
+MIT © [CarlosLeonCode](https://github.com/carlosleoncode) — see [LICENSE](LICENSE).

package/bin/cli.js

@@ -4,8 +4,16 @@
  * Invokes the main controller module for CLC Kernel.
  */
 
+const path = require('path');
 const { runCli } = require('../src/index');
 
+// Binary names from before the rename: still installed, removed in the next major.
+const DEPRECATED_BIN_NAMES = new Set(['clc-forge', 'create-clc-forge']);
+const invokedAs = path.basename(process.argv[1] || '').replace(/\.(js|cmd)$/i, '');
+if (DEPRECATED_BIN_NAMES.has(invokedAs)) {
+  process.stderr.write('"clc-forge" is deprecated; use "clckernel"\n');
+}
+
 runCli().catch((err) => {
   console.error('\n❌ Error ejecutando clckernel:', err);
   process.exit(1);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "clckernel",
-  "version": "1.2.18",
+  "version": "2.0.0",
   "description": "CLC Kernel — The Universal Polyglot AI Agent Governance Engine (Next.js, FastAPI, Django, Astro, Rails, Go, Rust, Laravel).",
   "main": "./bin/cli.js",
   "files": [
@@ -9,7 +9,9 @@
     "tools",
     ".agents/skills/clckernel_cli",
     "README.md",
-    "README.es.md"
+    "README.es.md",
+    "LICENSE",
+    "CHANGELOG.md"
   ],
   "bin": {
     "clckernel": "bin/cli.js",
@@ -41,6 +43,14 @@
   ],
   "author": "Carlos Leon Code",
   "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/CarlosLeonCode/clc_forge.git"
+  },
+  "bugs": {
+    "url": "https://github.com/CarlosLeonCode/clc_forge/issues"
+  },
+  "homepage": "https://clckernel.carlosleoncode.com",
   "engines": {
     "node": ">=18.0.0"
   },