verify-security — Claude Code security skill: checks and setup
The verify-security Claude Code skill checks code for secrets and security flaws. It runs on requests to review security, find secrets, or scan for vulnerabilities.
MIT ★ 44 updated 2026-10-13
npx skills add Aurite-ai/agent-verifier --skill verify-security -a claude-codeWhat verify-security does
The skill checks code for hardcoded secrets, unsafe handling of external input, exposed error details, and dependency version issues. It also reviews security-sensitive settings and labels each finding [P] for a pattern check or [H] for a heuristic judgment.
It examines configuration and source files while excluding tests and dependency folders. It never opens .env, .env.local, or .env.*.local; it only checks whether those files are ignored by Git. The skill makes no network calls or telemetry of its own. The host coding agent and its model provider process the code.
The output is a report with analyzed files, a summary, and findings. The instructions direct users to request “verify agent” for broader checks covering patterns, quality, and language-specific concerns.
What it helps with
- find API keys and passwords committed as string literals
- spot dependencies without pinned versions
- flag unsafe handling of user input and error responses
- catch insecure settings and sensitive values in logs
Example prompts for the agent
- Check this project for security flaws and exposed secrets.
- Scan the code for vulnerabilities and unsafe settings.
- Look for sensitive data in logs and details exposed by error responses.
How to install verify-security
1Skills CLI (Vercel)
npx skills add Aurite-ai/agent-verifier --skill verify-security -a claude-code2Manually: put the files in place
~/.claude/skills/verify-security/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/verify-security/
1Skills CLI (Vercel)
npx skills add Aurite-ai/agent-verifier --skill verify-security -a codex2Manually: put the files in place
.agents/skills/verify-security/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add Aurite-ai/agent-verifier --skill verify-security -a cursor2Manually: put the files in place
.cursor/skills/verify-security/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install Aurite-ai/agent-verifier skills/verify-security2Manually: put the files in place
.github/skills/verify-security/SKILL.md1Skills CLI (Vercel)
npx skills add Aurite-ai/agent-verifier --skill verify-security -a gemini-cli2Platform CLI
gemini skills install https://github.com/Aurite-ai/agent-verifier --path skills/verify-securityManually: put the files in place
.devin/skills/verify-security/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add Aurite-ai/agent-verifier --skill verify-security -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about verify-security
How do I install verify-security in Claude Code?
Run npx skills add Aurite-ai/agent-verifier --skill verify-security -a claude-code. The skill file is installed at ~/.claude/skills/verify-security/SKILL.md.
What does verify-security do?
It checks for hardcoded secrets, input validation problems, leaked error details, insecure settings, and dependency version issues. Findings are marked as pattern-based or heuristic.
Does verify-security open files containing real secrets?
No. It skips .env, .env.local, and .env.*.local files, checking only whether Git ignores them.
Source text
SKILL.md · MIT
---
name: verify-security
version: "1.0.0"
description: Verify code for security issues including hardcoded secrets, input validation, error exposure, and dependency vulnerabilities. Use when asked to "verify security", "check for secrets", or "scan for vulnerabilities".
---
# Security Verification
## Purpose
Verify code for security anti-patterns and vulnerabilities. No telemetry or network calls of its own; the code is processed by the host coding agent and its model provider.
## When to Use
Trigger this skill when the user asks to:
- "verify agent security"
- "verify security"
- "check for secrets"
- "scan for vulnerabilities"
- "security audit"
> **Note:** For full verification including patterns, quality, and language-specific checks, tell the user to say **"verify agent"**.
## Process
### Step 1: Discover Files
Locate files to analyze:
**Configuration files:**
- `package.json`, `pyproject.toml`, `Cargo.toml` - Dependencies
- `.env.example` - Environment template (flag values that look like real secrets)
- `.gitignore` - Confirm `.env` and `.env.local` are ignored (if not, flag it as an issue)
- `config.py`, `settings.py`, `config.ts` - Configuration
**Source files:**
- `*.py`, `*.ts`, `*.js`, `*.go`, `*.rs` - Source code
- Prioritize files with: `auth`, `api`, `client`, `secret`, `config` in name
**Exclude:**
- `node_modules/`, `.venv/`, `venv/`, `__pycache__/`
- `*.test.*`, `*.spec.*`, `*_test.go`
- `.env`, `.env.local`, `.env.*.local` - Never open these. They hold the user's real secrets. Check only whether they are gitignored.
### Step 2: Run Security Checks
#### Check Tiers
- **`[PATTERN]`** — Mechanical check. Apply exactly as written.
- **`[HEURISTIC]`** — Judgment required. Mark findings clearly.
Tag every finding with `[P]` for pattern or `[H]` for heuristic.
---
#### 2.1 `[PATTERN]` Hardcoded Secrets
Scan for assignments matching these patterns (case-insensitive):
| Variable pattern | Fail condition |
|------------------|----------------|
| `API_KEY` | Assigned to string literal |
| `SECRET` | Assigned to string literal |
| `PASSWORD` | Assigned to string literal |
| `TOKEN` | Assigned to string literal |
| `PRIVATE_KEY` | Assigned to string literal |
| `AWS_ACCESS_KEY_ID` | Assigned to string literal |
| `AWS_SECRET_ACCESS_KEY` | Assigned to string literal |
**Examples of failures:**
```python
# ❌ Issue
API_KEY = "sk-abc123..."
password = "hunter2"
OPENAI_API_KEY = "sk-proj-..."
# ✅ Pass (value loaded at runtime from env or a secrets manager, never a literal)
api_key = settings.api_key
password = secret_store.get("db-password")
```
**Also flag:**
- String literals matching known API key patterns:
- `sk-...` (OpenAI)
- `sk-ant-...` (Anthropic)
- `AKIA...` (AWS)
- `ghp_...` (GitHub)
- `xoxb-...` (Slack)
Severity: ❌ Issue
---
#### 2.2 `[PATTERN]` Dependency Version Pinning
**Python (`requirements.txt`):**
| Pattern | Severity |
|---------|----------|
| `package>=1.0` | ❌ Issue |
| `package>1.0` | ❌ Issue |
| `package` (no version) | ❌ Issue |
| `package==1.0.0` | ✅ Pass |
| `package~=1.0` | ✅ Pass |
**Python (`pyproject.toml`):**
Check `[project.dependencies]` and `[tool.poetry.dependencies]`:
- Unpinned or `>=` versions → ❌ Issue
- Pinned with `==` or `^` or `~` → ✅ Pass
**JavaScript/TypeScript (`package.json`):**
| Pattern | Severity |
|---------|----------|
| `"package": "*"` | ❌ Issue |
| `"package": "latest"` | ❌ Issue |
| `"package": ">=1.0.0"` | ⚠️ Warning |
| `"package": "^1.0.0"` | ✅ Pass |
| `"package": "~1.0.0"` | ✅ Pass |
| `"package": "1.0.0"` | ✅ Pass |
---
#### 2.3 `[HEURISTIC]` Input Validation
Check for external data handling:
**Look for:**
- HTTP request handlers (`@app.route`, `router.get`, etc.)
- User input processing (`request.body`, `req.params`, `input()`)
- File uploads
- Database queries with user input
**Flag if:**
- User input is passed directly to database queries without sanitization
- File paths are constructed from user input without validation
- JSON parsing without schema validation on external data
Severity: ⚠️ Warning
**Example patterns to flag:**
```python
# ⚠️ Warning - SQL without parameterization
query = f"SELECT * FROM users WHERE id = {user_id}"
# ⚠️ Warning - Path traversal risk
file_path = os.path.join(base_dir, user_filename)
# ✅ Pass - Parameterized query
cursor.execute("SELECT * FROM users WHERE id = ?", (user_id,))
```
---
#### 2.4 `[HEURISTIC]` Error Message Exposure
Check error handling for information leakage:
**Flag if:**
- Stack traces returned in HTTP responses
- Database error messages exposed to users
- Internal paths or system info in error messages
- Debug mode enabled in production code
**Look for:**
```python
# ⚠️ Warning
except Exception as e:
return {"error": str(e)} # Exposes internal details
# ⚠️ Warning
app = Flask(__name__)
app.debug = True # Debug in production
# ✅ Pass
except Exception as e:
logger.error(f"Error: {e}")
return {"error": "An error occurred"}
```
Severity: ⚠️ Warning
---
#### 2.5 `[HEURISTIC]` Secure Defaults
Check configuration for insecure defaults:
| Setting | Insecure | Secure |
|---------|----------|--------|
| CORS | `*` (allow all) | Specific origins |
| SSL verification | `verify=False` | `verify=True` or omitted |
| Debug mode | `debug=True` | `debug=False` |
| Cookie security | `secure=False` | `secure=True` |
| CSRF | Disabled | Enabled |
**Examples:**
```python
# ⚠️ Warning
requests.get(url, verify=False)
app.config["SESSION_COOKIE_SECURE"] = False
CORS(app, origins="*")
# ✅ Pass
requests.get(url) # verify=True is default
app.config["SESSION_COOKIE_SECURE"] = True
CORS(app, origins=ALLOWED_ORIGINS) # explicit allowlist
```
Severity: ⚠️ Warning
---
#### 2.6 `[HEURISTIC]` Sensitive Data Logging
Check logging statements for sensitive data:
**Flag if logging includes:**
- Passwords or tokens
- API keys
- Personal identifiable information (PII)
- Credit card numbers
- Session tokens
**Look for:**
```python
# ⚠️ Warning
logger.info(f"User login: {username} with password {password}")
print(f"API response: {response.json()}") # May contain tokens
# ✅ Pass
logger.info(f"User login: {username}")
logger.debug(f"Request to {url}") # No sensitive data
```
Severity: ⚠️ Warning
---
### Step 3: Generate Report
```markdown
# Security Verification Report
**Project:** [name or path]
**Date:** [current date]
**Files analyzed:** [count]
## Summary
✅ X checks passed | ⚠️ Y warnings | ❌ Z issues
## Secrets
- [x] No hardcoded secrets found
- [ ] ❌ Hardcoded secret at `[file:line]`
## Dependencies
- [x] All dependencies pinned
- [ ] ❌ Unpinned dependencies in `[file]`
## Input Validation
- [x] External input properly validated
- [ ] ⚠️ Potential injection at `[file:line]`
## Error Handling
- [x] Errors properly sanitized
- [ ] ⚠️ Information leakage at `[file:line]`
## Findings
> `[P]` = pattern-matched · `[H]` = heuristic
### ✅ Passing
- `[P]` No hardcoded API keys or secrets
- `[P]` Dependencies properly pinned
### ⚠️ Warnings
- `[H]` [Check]: [description]
- **Location:** [file:line]
- **Risk:** [what could go wrong]
- **Suggestion:** [how to fix]
### ❌ Issues
- `[P]` [Check]: [description]
- **Location:** [file:line]
- **Rule:** [which rule violated]
- **Fix:** [specific remediation]
## Recommendations
1. [Priority recommendation]
2. [Additional improvements]
```
---
*For full verification including patterns, quality, and language-specific checks, say "verify agent".*
© Aurite-ai · MIT · Aurite-ai/agent-verifier