Security

Navin AppSec security agent: /fortify, scanners and source → sink evidence

August 5, 2026 · 3 min read · Navin team

Local AppSec audit: surface map, scanners, injection, authz, supply chain, HTML report and numbered hardening plan.

Navin's security agent does what a serious AppSec audit requires: map the attack surface, cross-check scanners, trace source → sink with evidence, then deliver an HTML report and a numbered hardening plan. Read-only by default. You pick #1, the agent fixes.

How to start an audit

Three doors, one engine:

Security mode + free text (prefixed as /fortify).
/fortify [path|scope], or Actions → Security audit.
Security family: /probe, /xray, /pentest, /recon, /threatmap, /dast, /redteam, /comply, and more.

The model role is security (Settings → Models → Task routing): you choose which model audits, independently from the one that codes.

Nine phases covered end to end

Phase	What the agent covers
Surface map	Languages, lockfiles, routes, GraphQL/WS/webhooks, auth, SQL, uploads, jobs, IaC/Docker/K8s, CI secrets, admin panels
Scanners	gitleaks, trufflehog, npm audit, pip-audit, osv-scanner, bandit, semgrep, trivy, checkov, tfsec, kube-linter… when present
Injection and data	SQL/NoSQL/ORM, command/LDAP/template injection, path traversal, XXE, deserialization, SSRF
Frontend and client	XSS, `dangerouslySetInnerHTML`, open redirects, CSRF, CSP, postMessage, prototype pollution, client-only authz
AuthN / AuthZ	Hashing, JWT `alg:none`, sessions, MFA, IDOR/BOLA, escalation, mass assignment
Network and transport	TLS, HSTS, CORS, headers, rate limits, websocket auth, webhook signatures, exposed admin ports
Supply chain and secrets	Lockfile CVEs, typosquatting, keys in the tree or history, secrets in logs
Privacy and compliance	PII flows, encryption, retention, tenant isolation, GDPR/CCPA/PCI gaps
LLM / agent	Prompt injection, over-scoped tools, sandbox escape (when the project is agentic)

If a phase is clean, the agent says so with evidence, not with silence.

Every finding = real proof

Severity, file:line or scanner proof, impact, real PoC (payload, curl, code excerpt), minimal fix. No generic checklist. The dedicated security_scan tool (kinds secrets|sast|sca|quick|full) writes security-report-[date].html and opens it in File Preview.

Close: report + numbered choices

HTML report: executive summary, severity counters, finding cards, plan #1, #2… (effort, risk if delayed, first step).
Short chat: "Which number do we start with?".
Read-only unless you ask to fix.
On an open PR: pr_comments(kind=security) via gh.

Ethics and scope

Stay in authorized scope. Non-destructive PoCs. Never exfiltrate real secrets. You remain responsible for auditing only what you are allowed to audit.

Security + autonomous board

After /fortify, each confirmed finding can become a board task (status=fix). With board autonomy: isolated branch, green tests, PR, GitHub issue closed. The audit is no longer a forgotten PDF - it is the start of the pipeline.

FAQ

Does the agent really exploit my app?

It stays defensive and non-destructive by default. Offensive workflows (/pentest, /redteam, /dast) stay inside your authorized perimeter.

Does it replace a human pentester?

No. It accelerates continuous coverage and remediation. Business and legal decisions stay human.

What about secrets it finds?

They appear in the local report. Navin does not send them anywhere: your keys and findings stay on your machine.

Download Navin · Features · Documentation

Conclusion

A local AppSec audit with scanners, source → sink evidence, an HTML report and a numbered plan: that is the difference between a marketing "security scan" and an agent that lasts in your pipeline.

Try Navin on your machine

Local agent, cross-platform. Code, debug, scrape, leads, security and review - without leaving Navin.

Download See features