Security Agent Manual
Everything you need to know about your 24/7 AI security officer — how to read your security score, monitor vulnerabilities, run access reviews, scan for secrets, and manage pentest engagements.
1. What the Security Agent does
The Security Agent is your always-on AI security officer. It monitors your systems for vulnerabilities, unauthorized access, data exposures, and compliance drift — and acts on findings automatically where it safely can. Everything else it escalates to you with a clear recommended action.
Unlike a one-time penetration test, the Security Agent watches continuously. New CVEs that affect your stack are detected within hours of disclosure. Access reviews happen on schedule, not when an auditor demands them. Secrets accidentally committed to repositories are caught before they circulate.
What it monitors
| Domain | What is checked |
|---|---|
| Vulnerability scanning | CVEs in your dependencies, container images, and OS packages |
| Security events | Anomalous login patterns, privilege escalations, unusual API access |
| Access reviews | Who has access to what — and whether they still need it |
| Compliance | SOC 2 Type II, ISO 27001, PCI DSS, and custom policy checks |
| Secret scanning | API keys, tokens, and credentials in code, logs, and environment variables |
| Attack surface | Public-facing ports, subdomains, open S3 buckets, misconfigured CORS |
| DNS monitoring | DNS record changes, subdomain takeover risk, DNSSEC validity |
| Threat intelligence | Your domains and IPs against known threat feeds and dark web mentions |
2. Security score explained
Your security score is a 0–100 number calculated every hour. It reflects the overall posture of your systems based on open vulnerabilities, compliance gaps, access review status, and recent security events. A higher score means lower risk.
Score breakdown
| Component | Weight | What affects it |
|---|---|---|
| Vulnerability posture | 35% | Open CVEs by severity; age of unpatched vulns; patch velocity |
| Access hygiene | 25% | Overprovisioned accounts; stale access not reviewed; MFA coverage |
| Compliance coverage | 20% | Controls passing vs failing; time since last audit |
| Secret hygiene | 10% | Detected secrets; time to rotation after detection |
| Attack surface | 10% | Open ports; public buckets; misconfigured DNS; dangling subdomains |
Score thresholds
Strong posture. No critical gaps.
Minor gaps. Keep on top of them.
Meaningful risks. Action required.
Serious exposure. Address immediately.
Ask the agent for a score explanation at any time: "Why is our security score 67? What are the top 3 things I can fix to raise it above 80?"
3. Vulnerability and CVE monitoring
The agent continuously monitors the National Vulnerability Database (NVD) and major advisory feeds for CVEs affecting your stack. When a new CVE is published that matches a library or package you use, you receive an alert within hours — before public exploits are weaponized.
How to register your stack
"Monitor these for CVEs: Node.js 20.x, Next.js 14, PostgreSQL 15, Redis 7, Docker 24. Also scan our package.json at github.com/myorg/myapp — the GitHub token is ghp_abc123. Alert me immediately on CVSS ≥ 8.0, and weekly summary for anything below that."
CVE alert tiers
| CVSS Score | Severity | Alert timing | What the agent does |
|---|---|---|---|
| 9.0–10.0 | Critical | Immediate (email + Slack) | Opens P0 ticket; suggests mitigation steps |
| 7.0–8.9 | High | Within 1 hour | Opens P1 ticket; assesses exploitability in your environment |
| 4.0–6.9 | Medium | Daily digest | Lists in weekly report; tracks patch status |
| 0.1–3.9 | Low | Weekly digest | Tracked; low urgency |
Getting a vulnerability report
4. Security event detection
The Security Agent monitors your authentication logs, API access patterns, and infrastructure events for anomalous behavior. It uses a baseline built from your normal traffic patterns and flags deviations that warrant investigation.
Events the agent detects
| Event type | What triggers it | Default severity |
|---|---|---|
| Brute force login | > 10 failed logins from the same IP in 5 minutes | High |
| Credential stuffing | Many failed logins across many accounts from rotating IPs | Critical |
| Impossible travel | Same user logs in from two locations too far apart in time | High |
| Privilege escalation | User granted admin rights outside of change control window | Critical |
| Unusual API access | Bulk data export or unusual endpoint sequence | Medium |
| Off-hours admin action | Admin action at 3 AM when team is in UTC+0 | Medium |
| New IP for privileged user | Admin account accessed from a new country for the first time | High |
Connecting log sources
The agent ingests logs from your auth provider, application, and infrastructure. Tell it where your logs live:
5. Access reviews
Access reviews verify that every person with access to a system still needs that access. The agent runs reviews on a schedule you define and generates a structured report showing who has access to what, when they last used it, and whether the access looks appropriate.
Setting up access reviews
What a review report includes
| Section | Content |
|---|---|
| Summary | Total accounts reviewed; action required count; clean count |
| Inactive accounts | Accounts with no activity in the review period — candidates for removal |
| Overprovisioned accounts | Accounts with permissions broader than their role requires |
| New accounts | Accounts created since the last review — confirm each was intentional |
| Admin accounts | Full list of admin/superuser accounts — verify each is still needed |
| Shared credentials | Accounts shared between multiple people — flag for individual account creation |
| Recommended actions | Numbered list of accounts to deactivate, demote, or investigate |
Acting on a review
Review the report and tell the agent what to do: "Deactivate the AWS IAM user legacy_deploy — it has been inactive for 180 days and is no longer needed. Also demote user contractor_alice from admin to read-only on the GitHub organization."
6. Compliance tracking
The agent tracks your compliance against security frameworks and company policies. It maps your current controls to framework requirements and tells you exactly which controls are passing, failing, or not yet implemented.
Supported frameworks
| Framework | What the agent checks |
|---|---|
| SOC 2 Type II | CC trust service criteria: security, availability, confidentiality, processing integrity, privacy |
| ISO 27001:2022 | Annex A controls — 93 controls across 4 themes |
| PCI DSS v4.0 | 12 requirements for cardholder data environments |
| HIPAA Security Rule | Administrative, physical, and technical safeguards |
| Custom policy | Upload your own security policy document; agent checks against it |
Getting a compliance report
Continuous compliance monitoring
Rather than point-in-time audits, the agent monitors compliance continuously. When a configuration change causes a control to fail — for example, disabling MFA on an admin account — you receive an alert immediately rather than discovering it during an audit six months later.
7. Secret scanning
Leaked credentials are the most common way attackers gain initial access. The agent scans your repositories, CI/CD logs, and environment configurations for accidentally exposed secrets — API keys, tokens, private keys, database URLs, and passwords.
How to set up secret scanning
What the agent looks for
| Pattern | Examples detected |
|---|---|
| Cloud provider keys | AWS access keys (AKIA...), GCP service account JSON, Azure connection strings |
| Payment keys | Stripe sk_live_..., PayPal production client secrets |
| Database URLs | postgres://user:password@host/db, mongodb+srv://... |
| API tokens | GitHub PATs, Slack bot tokens, Twilio auth tokens, SendGrid keys |
| Private keys | RSA/EC private keys (-----BEGIN PRIVATE KEY-----) |
| JWT secrets | Hard-coded signing secrets used for JWT verification |
| Custom patterns | Any regex pattern you provide — e.g. your internal API key format |
Secret rotation workflow
When a secret is found, tell the agent: "I have rotated the Stripe key. The new key starts with sk_live_Abc. Update all references you know about and confirm the old key is gone from the codebase."
8. Attack surface and DNS monitoring
Your attack surface is everything an attacker can reach from the outside: open ports, exposed admin panels, public cloud storage, misconfigured CORS, and forgotten subdomains. The agent maps and monitors all of it.
What gets monitored
| Asset type | What the agent checks |
|---|---|
| Ports and services | Open ports that should be closed; services running on non-standard ports |
| Subdomains | New subdomains (could be rogue), subdomains pointing to deleted services (takeover risk) |
| S3 / GCS / Azure Blob | Public buckets containing sensitive data; directory listing enabled |
| CORS configuration | Wildcard CORS on APIs that handle authentication; credentials=true misuse |
| Admin panels | Publicly accessible /admin, /phpmyadmin, /.env, /.git paths |
| TLS / SSL | Expired or soon-to-expire certificates; weak cipher suites; HSTS not set |
| DNS records | MX, SPF, DKIM, DMARC — email spoofing protection |
Subdomain takeover risk
Subdomain takeover happens when a DNS CNAME record points to a cloud service (Heroku, GitHub Pages, S3, Vercel) that no longer exists. An attacker can claim the deleted service and receive traffic intended for your subdomain. The agent detects these automatically and alerts you to remove or fix the dangling record.
Setting up attack surface monitoring
9. Threat intelligence
The agent monitors threat intelligence feeds to check whether your domains, IPs, or employee emails have appeared on blocklists, in breach databases, or in dark web marketplaces. Knowing before your customers or regulators is what matters.
What is checked
| Feed type | What it tells you |
|---|---|
| IP reputation | Your outbound IPs on spam or abuse blocklists — email deliverability and firewall risk |
| Domain blocklists | Your domains flagged as malicious or phishing by security vendors |
| Breach databases | Employee email addresses found in known credential dumps |
| Dark web mentions | Your company name, domain, or brand mentioned in hacker forums or marketplaces (Studio tier) |
| Phishing detection | Lookalike domains (myc0mpany.com) registered that could be used to impersonate you |
Responding to a breach database hit
1. Force a password reset for the affected account immediately
2. Check whether the same email/password combination was used for internal systems
3. Review recent activity on the affected account for signs of compromise
Ask the agent: "User alice@mycompany.com appears in the breach dump. Guide me through the investigation."
10. Pentest management
The Security Agent manages your penetration testing engagements end-to-end — from scoping and scheduling through to tracking remediation of findings. It does not perform the pentest itself (that requires human security researchers), but it coordinates the entire process and tracks all findings as structured tickets.
Starting a pentest engagement
Finding lifecycle
| Status | Meaning |
|---|---|
| New | Finding received from pentest team; not yet assessed by internal team |
| Triaged | Internal team confirmed validity and assigned owner and due date |
| In remediation | Fix is being worked on |
| Fixed | Fix deployed — awaiting retest by pentest team |
| Verified | Pentest team confirmed the fix is effective |
| Accepted risk | Finding acknowledged but accepted due to low impact or compensating controls |
| False positive | Finding assessed and confirmed not exploitable in your environment |
Tracking remediation
Every finding is created as a structured ticket and tracked like any other security ticket. The agent monitors remediation progress and alerts you when findings approach their due date unresolved. After fixes are deployed, it notifies the pentest team for retest.
Executive report
At the end of a pentest engagement, ask: "Generate an executive summary for the July pentest engagement. Include: total findings by severity, how many are remediated, how many are outstanding, and what the overall risk posture improvement looks like."
11. Communicating with your agent
Your Security Agent session has a dedicated email (security-a3f9b2c1@agents.boboyka.com) and a live chat panel at hire.boboyka.com/session/[id]. Both share the same conversation history.
Things you can ask the agent
| Request | Example |
|---|---|
| Get current score | "What is our security score right now and what is dragging it down?" |
| Investigate an event | "There was a login from IP 203.0.113.42 at 3 AM — is this suspicious?" |
| Scan for secrets | "Run a secret scan on our monorepo right now — we just onboarded three new engineers" |
| Run an access review | "Run an access review for our AWS account — focus on IAM roles with admin permissions" |
| Check compliance | "Are we currently passing all SOC 2 Security criteria?" |
| Add a pentest finding | "Add finding: stored XSS on /profile/bio, CVSS 7.5, medium severity" |
| Rotate a secret | "We found our Resend API key in a public commit. Guide me through rotation." |
| Check attack surface | "Did any new subdomains appear on our domain in the last 24 hours?" |
12. Tier capabilities
| Feature | Starter | Pro | Studio |
|---|---|---|---|
| Security score | ✅ (daily) | ✅ (hourly) | ✅ (real-time) |
| CVE monitoring | ✅ (public packages) | ✅ (+ private repos) | ✅ (+ container images) |
| Security event detection | ✅ (basic patterns) | ✅ (full anomaly detection) | ✅ (ML-based) |
| Access reviews | ❌ | ✅ (quarterly) | ✅ (any schedule) |
| Compliance tracking | ❌ | ✅ (SOC 2) | ✅ (all frameworks) |
| Secret scanning | ✅ (public repos) | ✅ (all repos + CI) | ✅ (+ runtime env) |
| Attack surface monitoring | ✅ (basic ports + TLS) | ✅ (full surface) | ✅ (+ dark web) |
| Threat intelligence | ❌ | ✅ (IP + domain) | ✅ (+ dark web + phishing) |
| Pentest management | ❌ | ✅ | ✅ (+ executive reports) |
| Slack alerts | ❌ | ✅ | ✅ |
| Agent chat panel | ✅ | ✅ | ✅ |
13. Troubleshooting
My security score dropped significantly overnight
Ask the agent directly: "Our score dropped from 78 to 61 overnight. What changed?" The agent will show you the exact factors that changed — typically a new critical CVE was published, a compliance control started failing, or a new finding was classified as high severity.
The agent found a secret — what do I do right now?
2. Check the repository's git history — the secret may exist in older commits even after you delete it from the latest version. Rewrite history or consider the repo compromised.
3. Check your access logs for the compromised credential — look for any usage you did not authorize.
4. Tell the agent: "I have rotated the key. Verify the old key is gone and confirm no unauthorized usage in the last 30 days."
A pentest finding is marked false positive but the agent keeps tracking it
Tell the agent: "Mark pentest finding PT-024 as false positive. Reason: the CORS wildcard is intentional — the endpoint only serves public data and has no authentication." The agent records the reason and removes it from the open findings count. The finding is retained in the audit trail.
Access review is flagging contractors as inactive
Contractors often have lower activity during off-project periods but still need persistent access. Tell the agent: "Exclude users with the GitHub label 'contractor' from the inactivity threshold — use a 180-day threshold instead of 90 days for contractors."
I received a threat intel alert about my domain on a blocklist
This usually means one of your servers sent spam or was used in an attack. Ask the agent: "Our domain mycompany.com is on the Spamhaus blocklist. Which of our IPs caused this and how do I get delisted?" The agent will identify the source and provide the delisting steps for the specific blocklist.