[codex] Clarify CSReview scope and add GitHub security validation

.github/dependabot.yml

@@ -0,0 +1,22 @@
+version: 2
+
+updates:
+  - package-ecosystem: npm
+    directory: /csreview
+    schedule:
+      interval: weekly
+      day: monday
+      time: "06:00"
+      timezone: America/Sao_Paulo
+    groups:
+      npm-runtime:
+        patterns:
+          - "*"
+
+  - package-ecosystem: github-actions
+    directory: /
+    schedule:
+      interval: weekly
+      day: monday
+      time: "06:30"
+      timezone: America/Sao_Paulo

.github/workflows/ci.yml

@@ -0,0 +1,49 @@
+name: CI
+
+on:
+  pull_request:
+  push:
+    branches:
+      - main
+
+permissions:
+  contents: read
+
+jobs:
+  test:
+    name: Node ${{ matrix.node-version }}
+    runs-on: ubuntu-latest
+    timeout-minutes: 15
+    strategy:
+      fail-fast: false
+      matrix:
+        node-version:
+          - 20
+          - 24
+
+    defaults:
+      run:
+        working-directory: csreview
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v5
+
+      - name: Setup Node.js
+        uses: actions/setup-node@v5
+        with:
+          node-version: ${{ matrix.node-version }}
+          cache: npm
+          cache-dependency-path: csreview/package-lock.json
+
+      - name: Install dependencies
+        run: npm ci
+
+      - name: Run tests
+        run: npm test
+
+      - name: Audit dependencies
+        run: npm audit --audit-level=low
+
+      - name: Verify package contents
+        run: npm pack --dry-run

.github/workflows/semgrep.yml

@@ -0,0 +1,43 @@
+name: Semgrep
+
+on:
+  pull_request:
+  push:
+    branches:
+      - main
+  schedule:
+    - cron: "47 3 * * 1"
+
+permissions:
+  contents: read
+  security-events: write
+
+jobs:
+  semgrep:
+    name: Semgrep SARIF
+    runs-on: ubuntu-latest
+    timeout-minutes: 20
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v5
+
+      - name: Install Semgrep
+        run: python -m pip install --upgrade semgrep
+
+      - name: Run Semgrep
+        continue-on-error: true
+        run: |
+          semgrep scan \
+            --config auto \
+            --sarif \
+            --output semgrep.sarif \
+            --exclude node_modules \
+            --exclude csreview-reports \
+            .
+
+      - name: Upload Semgrep SARIF
+        uses: github/codeql-action/upload-sarif@v4
+        if: always() && hashFiles('semgrep.sarif') != ''
+        with:
+          sarif_file: semgrep.sarif

README.md

@@ -6,7 +6,7 @@
 
 ## Purpose & Utility
 
-**CSReview** is a universal AI agent skill that performs **ultra-deep security audits** (automated pentest-style analysis) on codebases across multiple languages, frameworks, and platforms.
+**CSReview** is a universal AI agent skill for development-time security alignment of the local workspace a developer is actively building. It applies a penetration tester's adversarial mindset to local source, configuration, dependencies, and infrastructure files (static SAST + SCA), but it does not perform live penetration testing against running, deployed, or production systems.
 
 ### CSReview is READ-ONLY
 
@@ -18,7 +18,14 @@ CSReview **NEVER modifies, deletes, or moves any files** in the analyzed project
 
 The actual fixes are applied by the human developer or the coding agent, not by CSReview.
 
-CSReview exists to slow down unsafe "vibe coding" before production: it inspects code, dependency manifests, framework configuration, database/BaaS rules, frontend/backend boundaries, and platform-specific surfaces, then writes a detailed report explaining what is exposed and why. It does not assume enough business or schema context to change the audited system itself.
+CSReview exists to slow down unsafe "vibe coding" before release: it inspects local code, dependency manifests, framework configuration, database/BaaS rules, frontend/backend boundaries, and platform-specific surfaces, then writes a detailed report explaining what is exposed and why. It does not assume enough business or schema context to change the audited system itself.
+
+### Scope
+
+- **IN SCOPE**: the local development workspace/project, including local source code, configuration, `.env` files, infrastructure-as-code, and BaaS rule files. Local SAST/SCA tools such as Semgrep, npm audit, OSV-Scanner, and framework-native scanners may be run against that local code only.
+- **GOAL**: improve the SECURITY and EFFICIENCY (cost/performance) of the project under development.
+- **OUT OF SCOPE / PROHIBITED**: testing, probing, or calling live, deployed, or production systems; external service endpoints used by the app; DAST against running targets; modifying audited code; exfiltrating data.
+- **Reference documentation research is ALLOWED**: reading OWASP, CWE, CVE/NVD, OSV, vendor advisories, and official framework documentation to ground remediation is allowed. That is reading documentation, not probing a target.
 
 ### Global Skill Installation Only
 
@@ -77,7 +84,7 @@ The verbal summary is not enough for implementation. A coding agent must analyze
 
 Security vulnerabilities cost companies billions annually. Most development teams lack dedicated security engineers to review code before deployment. With the rise of **vibe coding** (non-technical users building software with AI agents), security risks have multiplied. CSReview bridges this gap by providing:
 
-1. **Automated Pentest-Level Analysis**: Goes beyond basic linting - performs the same depth of analysis a human security consultant would do
+1. **Development-Time Security Alignment**: Goes beyond basic linting with static source/config review, Semgrep/SCA evidence, and a security consultant's adversarial reasoning without probing live systems
 2. **Dual Report System**:
    - **HTML Report** in the user's language for human understanding
    - **Markdown Report** in English for AI coding agents to parse and plan remediations without changing the audited code automatically
@@ -91,7 +98,7 @@ Security vulnerabilities cost companies billions annually. Most development team
 
 ### Real-World Use Cases
 
-- **Pre-deployment security gate**: Run before pushing to production
+- **Pre-release security gate**: Run before release while reviewing only the local workspace
 - **Code review enhancement**: Augment human code reviews with automated security analysis
 - **Legacy code audit**: Identify vulnerabilities in existing codebases
 - **Compliance preparation**: Find issues before security audits (SOC 2, ISO 27001, LGPD/GDPR)
@@ -417,7 +424,7 @@ Simply ask your AI coding assistant:
 
 ### Pre-Deployment Review
 ```
-@csreview Run a full security audit before production deployment
+@csreview Run a full local workspace security review before release
 ```
 
 ### Backend Security

SECURITY.md

@@ -1,49 +1,59 @@
 # Security Policy
 
-CSReview is a white-hat, development-time security tool. We take the security of
-the tool itself — and of the reports it generates — seriously.
+CSReview is a white-hat, development-time security alignment tool for local
+workspaces. We take the security of the tool itself, its instructions, and the
+reports it generates seriously.
 
 ## Supported Versions
 
 Only the latest commit on the `main` branch is supported. Please update before
 reporting an issue.
 
-| Version              | Supported |
-| -------------------- | --------- |
-| `main` (latest)      | ✅        |
-| older commits / forks| ❌        |
+| Version               | Supported |
+| --------------------- | --------- |
+| `main` (latest)       | Yes       |
+| older commits / forks | No        |
 
 ## Reporting a Vulnerability
 
-**Please do not open public Issues or Pull Requests for security vulnerabilities.**
+Do not open public Issues or Pull Requests for security vulnerabilities.
 
-Report privately through GitHub's **Private Vulnerability Reporting**:
+Report privately through GitHub's Private Vulnerability Reporting:
 
 1. Open the [Security tab](https://github.com/decksoftware/csreview/security) of this repository.
-2. Click **"Report a vulnerability"**.
-3. Include the affected files/lines, the impact, and a reproduction if possible.
+2. Click "Report a vulnerability".
+3. Include the affected files/lines, impact, local reproduction steps, and relevant static evidence.
 
-You can expect an initial acknowledgement within **5 business days**. Validated
+Avoid including real secrets, tokens, private customer data, or exploit output.
+Redact sensitive values before sharing evidence.
+
+You can expect an initial acknowledgement within 5 business days. Validated
 issues are fixed on `main` and credited, unless you prefer to remain anonymous.
 
 ## Scope
 
-**In scope**
-- The CSReview skill code (`csreview/src/**`) and the CLI.
-- Security weaknesses in the **generated reports** — e.g., script injection into the
-  HTML report, or secret/credential leakage in the Markdown report.
-- The skill instructions (`SKILL.md`) where they could steer a coding agent into
-  unsafe behavior.
+In scope:
+
+- The CSReview skill code (`csreview/src/**`) and CLI.
+- Security weaknesses in generated reports, such as script injection into the
+  HTML report or credential leakage in the Markdown report.
+- Skill instructions (`SKILL.md`) that could steer a coding agent into unsafe
+  behavior.
+- Static-analysis behavior that violates the documented local-only, read-only
+  scope.
+
+Out of scope:
 
-**Out of scope**
-- Findings that CSReview reports about *your own* audited code — that is the tool
+- Findings that CSReview reports about your own audited code. That is the tool
   working as intended, not a vulnerability in CSReview.
-- Vulnerabilities in third-party scanners (Semgrep, OSV-Scanner, npm audit, etc.) —
-  please report those to their respective upstream projects.
-- Social engineering, or anything requiring access to a maintainer's machine.
+- Vulnerabilities in third-party scanners such as Semgrep, OSV-Scanner, npm
+  audit, CodeQL, or similar tools. Report those to their upstream projects.
+- Social engineering or anything requiring access to a maintainer's machine.
+- Testing, probing, or calling live systems, production services, external
+  application endpoints, or user data while reporting an issue here.
 
 ## Our Commitment
 
-CSReview is **read-only** on audited code and never tests live, deployed, or
+CSReview is read-only on audited code and never tests live, deployed, or
 production systems. If you find a way it violates that guarantee, treat it as a
 high-priority report.

csreview/SKILL.md

@@ -7,7 +7,14 @@ description: "Ultra-deep security audit and pentest analysis for codebases. Gene
 
 ## Overview
 
-This skill performs ultra-deep security analysis (automated pentest-style analysis) on codebases across multiple languages, frameworks, and platforms. It identifies vulnerabilities, data leakage risks, misconfigurations, and security flaws, then generates:
+This skill performs development-time security alignment for the local workspace a developer is actively building. It applies a penetration tester's adversarial mindset to the project's local source, configuration, dependencies, and infrastructure files (static SAST + SCA). It does NOT perform live penetration testing against running, deployed, or production systems. It identifies vulnerabilities, data leakage risks, misconfigurations, and security flaws, then generates:
+
+## Scope
+
+- **IN SCOPE**: the local development workspace/project, including all local source code, configuration, `.env` files, infrastructure-as-code, and BaaS rule files. Local SAST/SCA tools such as Semgrep, npm audit, OSV-Scanner, and framework-native scanners may be run against that local code only.
+- **GOAL**: improve the SECURITY and EFFICIENCY (cost/performance) of the project under development.
+- **OUT OF SCOPE / PROHIBITED**: testing, probing, or calling live, deployed, or production systems; external service endpoints used by the app; DAST against running targets; modifying audited code; exfiltrating data.
+- **Reference documentation research is ALLOWED**: reading OWASP, CWE, CVE/NVD, OSV, vendor advisories, and official framework documentation to ground remediation is allowed. That is reading documentation, not probing a target.
 
 1. **HTML Report** (`csreview-reports/<agent>_security-report.html`) - Visual report for human review with executive summary, charts, and detailed findings
 2. **Markdown Report** (`csreview-reports/<agent>_security-findings.md`) - Structured report for humans and coding agents to understand, prioritize, and plan remediations without CSReview modifying the audited code
@@ -51,7 +58,7 @@ CSReview includes built-in **Code Review** capabilities (equivalent to codex:rev
 - User asks for vulnerability scan or pentest analysis
 - User wants to check for data leakage or exposed secrets
 - User mentions SQL injection, XSS, auth flaws, or security concerns
-- Before production deployment for security validation
+- Before release or deployment preparation, while reviewing only the local workspace
 - User asks to review Supabase, Firebase, Appwrite, Neon, or similar backend security
 - User invokes `@csreview` or mentions CSReview
 - User wants compliance verification (LGPD, GDPR, SOC 2, HIPAA)
@@ -1214,8 +1221,8 @@ For each vulnerability found:
 │ Vulnerable Code:                                │
 │ [syntax-highlighted code snippet]               │
 │                                                 │
-│ Exploitation Scenario:                          │
-│ Step-by-step how an attacker could exploit      │
+│ Potential Exploitation Path (theoretical, unverified): │
+│ Static-analysis hypothesis; not a validated or executed exploit. │
 │                                                 │
 │ Impact:                                         │
 │ What could happen if exploited                  │
@@ -1299,8 +1306,8 @@ For each vulnerability found:
           <p>...</p>
           <h3>Vulnerable Code</h3>
           <pre><code class="language-typescript">...</code></pre>
-          <h3>Exploitation Scenario</h3>
-          <p>...</p>
+          <h3>Potential Exploitation Path (theoretical, unverified)</h3>
+          <p>Static-analysis hypothesis only; not a validated or executed exploit.</p>
           <h3>Impact</h3>
           <p>...</p>
           <h3>Recommended Fix</h3>
@@ -1419,7 +1426,9 @@ const query = `SELECT * FROM users WHERE email = '${email}' AND password = '${pa
 const result = await db.query(query);
 ```
 
-#### Exploitation Scenario
+#### Potential Exploitation Path (theoretical, unverified)
+This is a hypothesis derived from static analysis, not a validated or executed exploit.
+
 1. Attacker submits email: `' OR '1'='1`
 2. Query becomes: `SELECT * FROM users WHERE email = '' OR '1'='1' AND password = 'anything'`
 3. Authentication bypass achieved

csreview/package.json

@@ -1,21 +1,21 @@
 {
   "name": "csreview",
-  "version": "1.0.0",
+  "version": "0.0.1",
   "description": "Ultra-deep security audit and pentest analysis for codebases",
   "main": "src/index.js",
   "bin": {
     "csreview": "./src/cli.js"
   },
   "type": "module",
   "engines": {
-    "node": "20 || >=22"
+    "node": ">=18"
   },
   "scripts": {
-    "test": "node --test \"test/**/*.test.js\"",
+    "test": "node --test test/analysis.test.js",
     "doctor": "node src/cli.js --doctor"
   },
   "keywords": ["security", "audit", "pentest", "code-review"],
-  "author": "",
+  "author": "decksoftware",
   "license": "MIT",
   "files": [
     "src/",