BridgeSecurity

<p align="center"> <strong>BridgeSecurity</strong> </p> <h3 align="center">Find vulnerabilities. Ship secure.</h3> <p align="center"> A Claude Code plugin from <a href="https://www.bridgemind.ai">BridgeMind</a> that gives your AI agents the instincts of a senior application security engineer.<br/> Stop shipping classic vulnerabilities — start shipping production-secure code. </p> <p align="center"> <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue.svg" alt="MIT License" /></a> <a href="https://www.bridgemind.ai/discord"><img src="https://img.shields.io/badge/Discord-builders-5865F2?logo=discord&logoColor=white" alt="Discord" /></a> </p>

---

Why BridgeSecurity?

AI coding agents write functional code, but they keep shipping the same classic vulnerabilities — SQL injection, XSS, IDOR, hardcoded secrets, missing auth on Server Actions, public S3 buckets, pull_request_target with checkout-of-fork-code. The bugs that have headlined CVEs for fifteen years.

BridgeSecurity fixes this. It's a set of detection patterns, vulnerability taxonomies, threat-modeling discipline, and a specialized auditor agent that teach your AI teammates to think like a senior security engineer — find the trust boundary, match input to sink, check auth on every state-changing path, treat every secret as already leaked, fail closed.

The Bugs It Catches

• Injection — SQLi, NoSQLi, command, code, template, XSS, LDAP, XPath
• Broken Access Control — IDOR, BOLA, BOPLA, missing auth, mass assignment, Server Action authz gaps, x-middleware-subrequest bypass class
• Cryptographic Failures — MD5/SHA1, Math.random() for tokens, ECB mode, hardcoded keys, JWT alg: none, HS256/RS256 confusion, === for HMAC compare
• SSRF — including private-IP / cloud-metadata / file:// / TOCTOU
• Path Traversal — path.join without prefix-check, send_file with raw input
• Insecure Deserialization — pickle.loads, yaml.load, ObjectInputStream, vm2
• Hardcoded Secrets — AWS, GitHub, Stripe, OpenAI, Anthropic, Slack, JWTs, private keys (25+ patterns)
• Cloud Misconfig — public S3, IAM *:*, 0.0.0.0/0:22, IMDSv1, missing encryption
• Container / k8s — privileged: true, runAsUser: 0, hostNetwork, /var/run/docker.sock mount, image:latest
• CI/CD — pull_request_target + checkout-fork-code, mutable Action tags (CVE-2025-30066 class), shell-injection via PR title
• Auth flaws — JWT in localStorage, no MFA, missing rate limit on login, predictable reset tokens
• Open Redirect — //evil.com bypass class
• Markdown exfil — image URLs constructed from secrets (EchoLeak class)
• LLM-specific — output-to-eval, excessive agency, secrets in system prompt

---

What's Inside

| Component | Type | What It Does | |-----------|------|-------------| | bridgesecurity | Skill | Core security discipline — auto-loaded when your agent reads, writes, or reviews code. Five Disciplines, threat-model checklist, detection cheat-sheet. | | security-audit | Skill | Slash-command audit. Scans a file/dir/PR/repo for vulnerabilities, returns severity-ranked report with CWE/OWASP mapping. | | security-auditor | Agent | Read-only senior security engineer subagent. Cannot write, edit, or delete. Walks every file with the OWASP Top 10 + CWE Top 25 + threat model. |

Reference Documentation

The skill ships with eight deep reference docs (~50 pages of practitioner-grade content):

• vulnerability-taxonomies.md — OWASP Top 10 (2021), API Top 10 (2023), Mobile Top 10 (2024), LLM Top 10 (2025), CWE Top 25 (2024), CISA KEV recurring classes, DBIR top vectors
• language-patterns.md — JS/TS, Python, Go, Rust, Java/Spring, Ruby/Rails, PHP — vulnerable + fixed code pairs, per-ORM SQLi reference
• frontend-patterns.md — React, Next.js (Server Actions, middleware, hydration), Vue, Svelte, browser specifics
• infrastructure-patterns.md — AWS, GCP, Azure, Docker, Kubernetes, Terraform, GitHub Actions, GitLab CI
• secrets-patterns.md — regex catalog for 25+ secret types
• case-studies.md — Log4Shell, Spring4Shell, MOVEit, XZ backdoor, Polyfill.io, Snowflake, Ivanti, regreSSHion, Next.js CVE-2025-29927, tj-actions supply chain, recent agent CVEs
• tooling.md — Semgrep, CodeQL, Snyk, Trivy, Gitleaks, OSV, govulncheck, Brakeman, Checkov, kube-bench, MobSF
• threat-modeling.md — the full 10-question discipline

---

Install

As a Claude Code plugin

claude plugin install bridgesecurity@bridgemind-plugins

Or copy the skills manually

# Project-level
mkdir -p .claude/skills .claude/agents
cp -r skills/bridgesecurity .claude/skills/
cp -r skills/security-audit .claude/skills/
cp agents/security-auditor.md .claude/agents/

# Personal / global
mkdir -p ~/.claude/skills ~/.claude/agents
cp -r skills/bridgesecurity ~/.claude/skills/
cp -r skills/security-audit ~/.claude/skills/
cp agents/security-auditor.md ~/.claude/agents/

---

How It Works

The Five Disciplines

Once installed, the bridgesecurity skill activates whenever your agent reads, writes, or reviews code. Your agent now thinks like a senior:

1. Find the trust boundaries — every place untrusted data crosses into trusted code.
2. Match input to sink — every (source, sink) pair is a potential vulnerability.
3. Auth on every state-changing path — authentication + authorization (ownership, not just role) + input validation.
4. Secrets are already leaked — assume rotation as default, plan for it.
5. Fail closed, log loudly, blast-radius small — default deny, sanitized errors, audit logs, network/IAM segmentation.

The 10-Question Threat Model

Before declaring code "secure":

1. Trust boundaries — where does data cross?
2. AuthN/AuthZ — server-side, ownership-checked?
3. Input validation — schema at the boundary?
4. Output encoding — correct context?
5. Secrets — safe storage + rotation?
6. Failure mode — fail closed?
7. Blast radius — what falls if owned?
8. Supply chain — pinned + audited?
9. Logging — security events captured? PII redacted?
10. Replay protection — idempotency, nonces, CSRF, rate limits?

If any answer is "I don't know" — the code is not cleared.

Audit Mode

> /security-audit src/api/users.ts

> /security-audit https://github.com/owner/repo/pull/123

> /security-audit ./terraform/

> /security-audit "all server actions in this app"

The security-auditor agent walks the target, applies all 20 audit categories, and produces a severity-ranked report with:

• File:line — exact location
• Class — vulnerability category
• CWE / OWASP — canonical IDs
• Code — verbatim vulnerable snippet
• Exploit scenario — concrete, not generic
• Fix — patched code
• References — links to CWE, OWASP, BridgeSecurity references

---

Why "BridgeSecurity"?

Direct. Functional. Pairs with BridgeWard — Trust nothing. Ship safely. — for prompt-injection defense:

• BridgeWard protects your agent from prompt injection in untrusted content.
• BridgeSecurity protects your code from classic application vulnerabilities.

Both rooted in the BridgeMind brand line: Ship with agents. The security corollary: Find vulnerabilities. Ship secure.

---

When to Use BridgeSecurity

You should install BridgeSecurity if your agent:

• Writes any production code
• Reviews PRs or diffs
• Generates infrastructure-as-code (Terraform, Kubernetes, Dockerfiles)
• Edits CI/CD configurations (GitHub Actions, GitLab CI)
• Refactors authentication / authorization / payments code
• Handles user input on any code path
• Writes Server Actions, RPC handlers, API endpoints
• Touches anything with eval, exec, query, fetch, redirect, file paths, or secrets

If your agent only writes pure functions and unit tests, you may not need this. Everyone else does.

---

Project Layout

BridgeSecurity/
├── .claude-plugin/
│   └── plugin.json
├── skills/
│   ├── bridgesecurity/
│   │   ├── SKILL.md
│   │   └── references/
│   │       ├── vulnerability-taxonomies.md
│   │       ├── language-patterns.md
│   │       ├── frontend-patterns.md
│   │       ├── infrastructure-patterns.md
│   │       ├── secrets-patterns.md
│   │       ├── case-studies.md
│   │       ├── tooling.md
│   │       └── threat-modeling.md
│   └── security-audit/
│       └── SKILL.md
├── agents/
│   └── security-auditor.md
└── scripts/
    └── scan.sh

---

Compatibility

BridgeSecurity uses the standard SKILL.md / agent package format supported by 30+ AI coding tools.

| Tool | Skills | Subagent | Notes | |------|--------|----------|-------| | Claude Code | ✅ | ✅ | Full plugin support | | Cursor | ✅ | — | Drop into .cursor/skills/ | | Windsurf | ✅ | — | Skill format | | OpenAI Codex | ✅ | — | Skill format | | Gemini CLI | ✅ | — | Skill format | | Cline / Roo Code | ✅ | — | Skill format | | GitHub Copilot | ✅ | — | Via .github/copilot-instructions.md reference | | Continue.dev | ✅ | — | Skill format | | Goose | ✅ | — | Skill format |

---

What BridgeSecurity Is Not

• Not a SAST tool. It's a thinking skill. Pair it with Semgrep / CodeQL / Snyk / Trivy — the skill knows how to invoke them and synthesize output.
• Not a guarantee. Security is layered. This skill makes your agent better; it doesn't make your code invulnerable.
• Not a replacement for human review on high-stakes flows (payments, auth, crypto, IAM).
• Not a runtime defense. It catches issues at code-time. WAFs, sandboxes, and observability live elsewhere.

It is one layer in your stack. Layer it with: SAST in CI, dependency scanning, secret scanning, container scanning, IaC scanning, DAST, runtime observability, and human security review for high-stakes changes.

---

Companion Skills

• BridgeWard — Prompt-injection defense. Trust nothing. Ship safely.
• BridgeUI — Senior frontend design instincts.
• BridgeRemotion — Remotion video-engineering expertise.
• BridgeMotion — MIT-licensed React video framework.

---

Authoritative References

This skill synthesizes guidance from:

• OWASP Top 10 (2021)
• OWASP API Security Top 10 (2023)
• OWASP Mobile Top 10 (2024)
• OWASP LLM Top 10 (2025)
• CWE Top 25 (2024)
• CISA Known Exploited Vulnerabilities
• Verizon DBIR
• NIST NVD
• CIS Benchmarks
• NSA/CISA Kubernetes Hardening Guide v1.2
• GitHub Actions Security Hardening
• OWASP Cheat Sheet Series

---

Contributing

PRs welcome — especially for new vulnerability patterns, fresh case studies, and per-language additions. See CONTRIBUTING.md.

When adding a new pattern: include a real-world citation (CVE, writeup, or CVSS score). When adding a new case study: name the vendor, date, vector, and remediation.

---

License

MIT. See LICENSE. True open source. No license traps. Ship freely.

---

About BridgeMind

BridgeMind is an agentic organization — AI agents are teammates, not tools. We build open-s