Security Policy¶

Reporting Security Vulnerabilities¶

If you discover a security vulnerability in Sentinel DV, please report it responsibly:

DO NOT open a public GitHub issue for security vulnerabilities.

Instead, please use GitHub Security Advisories: - Private report: https://github.com/kiranreddi/sentinel-dv/security/advisories/new - Security center: https://github.com/kiranreddi/sentinel-dv/security

Include: - Description of the vulnerability - Steps to reproduce - Potential impact - Suggested fix (if any)

We will respond within 48 hours and work with you to address the issue.

Security Model¶

Sentinel DV is designed with security as a core principle. See Architecture → Security for the complete security model.

Key Security Features¶

1. Read-Only Design - MCP tools only read verification data - No write, delete, or modify operations - Prevents accidental or malicious data corruption

2. Automatic Redaction - Credentials automatically removed from logs - File paths sanitized - Sensitive data filtered before exposure

3. Path Sandboxing - File access restricted to allowed paths - Path traversal attacks prevented - Symbolic link following disabled

4. Bounded Outputs - Response size limits prevent DoS - Pagination enforced for large datasets - Memory usage capped

5. Input Validation - All inputs validated with Pydantic schemas - SQL injection prevented (parameterized queries) - No arbitrary code execution

Supported Versions¶

We actively maintain security updates for:

Security patches are backported to the latest stable release only.

Security Best Practices¶

For Users¶

1. Configure Path Restrictions

# Restrict file access by configuring `artifact_roots` in `config.yaml`.
# The server validates that roots exist and are readable at startup.
python -m sentinel_dv.server --config config.yaml

2. Use Separate Databases

# Use separate `config.yaml` files with different `index.path` values.
python -m sentinel_dv.server --config ./config-prod.yaml  # Production
python -m sentinel_dv.server --config ./config-dev.yaml   # Development

3. Regular Updates

# Keep Sentinel DV up to date
pip install --upgrade sentinel-dv

4. Monitor Logs

# Run under your process manager and collect stdout/stderr logs.
python -m sentinel_dv.server --config config.yaml

For Developers¶

1. Validate All Inputs

from pydantic import BaseModel, Field

class SafeInput(BaseModel):
    run_id: str = Field(pattern=r"^[A-Za-z0-9_-]+$")
    page: int = Field(ge=1, le=1000)

2. Sanitize Outputs

def redact_credentials(text: str) -> str:
    """Remove credentials from text."""
    patterns = [
        r"(password|token|key|secret)[:=]\s*[^\s]+",
        r"Bearer\s+[A-Za-z0-9_-]+",
    ]
    for pattern in patterns:
        text = re.sub(pattern, r"\1: [REDACTED]", text, flags=re.IGNORECASE)
    return text

3. Limit Resource Usage

MAX_PAGE_SIZE = 100
MAX_RESPONSE_SIZE = 10 * 1024 * 1024  # 10MB

def safe_query(page_size: int):
    if page_size > MAX_PAGE_SIZE:
        raise ValueError(f"Page size exceeds maximum ({MAX_PAGE_SIZE})")
    # Execute query...

4. Use Parameterized Queries

# ✅ Good - parameterized
cursor.execute(
    "SELECT * FROM tests WHERE run_id = ?",
    (run_id,)
)

# ❌ Bad - SQL injection risk
cursor.execute(
    f"SELECT * FROM tests WHERE run_id = '{run_id}'"
)

Known Security Considerations¶

1. Local File Access¶

Risk: Sentinel DV indexes local files Mitigation: - Configure path access by setting artifact_roots in config.yaml. - Run with least-privilege user account - Use read-only file mounts in Docker

Example:

# Docker with read-only mount
docker run -v ./verification:/data:ro sentinel-dv:latest

2. Database Access¶

Risk: Index database contains indexed data Mitigation: - Restrict database file permissions (chmod 600) - Store in secure directory - Encrypt at rest if needed

Example:

# Secure database permissions
chmod 600 ./sentinel_db/index.db
chown sentinel-user:sentinel-group ./sentinel_db/index.db

3. MCP Communication¶

Risk: MCP protocol uses stdio communication Mitigation: - MCP servers run locally only (no network exposure) - Communication authenticated by AI client (Claude/Cline) - No external network access required

4. Log Parsing¶

Risk: Parsing untrusted log files Mitigation: - No code execution from logs - Regex patterns bounded - File size limits enforced - Timeout for parsing operations

Example:

MAX_FILE_SIZE = 100 * 1024 * 1024  # 100MB
PARSE_TIMEOUT = 30  # seconds

@timeout(PARSE_TIMEOUT)
def parse_log(file_path: str):
    if os.path.getsize(file_path) > MAX_FILE_SIZE:
        raise ValueError("File too large")
    # Parse...

Threat Model¶

Assets¶

• Verification data: Test results, logs, coverage
• System files: Configuration, database
• User credentials: If present in logs (auto-redacted)

Threats¶

Assumptions¶

• MCP server runs in trusted environment
• AI client (Claude/Cline) is trusted
• User has legitimate access to verification data
• Database is stored securely

Security Audits¶

Internal Reviews¶

• ✅ Code review for all changes
• ✅ Automated security scanning (Dependabot)
• ✅ SAST with Bandit and Ruff
• ✅ Dependency vulnerability scanning

External Audits¶

No external security audits have been performed yet. If you're interested in conducting a security audit, please contact us.

Vulnerability Disclosure Timeline¶

1. T+0: Vulnerability reported
2. T+48h: Acknowledgment sent
3. T+7d: Initial assessment completed
4. T+30d: Fix developed and tested
5. T+45d: Security release published
6. T+90d: Public disclosure (if agreed)

We follow responsible disclosure practices and will work with reporters to coordinate public disclosure.

Security Contacts¶

• Private security reports: https://github.com/kiranreddi/sentinel-dv/security/advisories/new
• Security advisories: https://github.com/kiranreddi/sentinel-dv/security/advisories
• General issues: https://github.com/kiranreddi/sentinel-dv/issues

Acknowledgments¶

We thank the following researchers for responsibly disclosing security issues:

(None yet - this section will be updated as security reports are received and addressed)

Related Documentation¶

• Architecture → Security Model - Complete security design
• Installation Guide - Secure installation practices
• Contributing Guide - Secure development practices

Updates¶

This security policy was last updated: January 25, 2026

Security policy changes are tracked in CHANGELOG.md.