Secure SDLC Agents

8 AI security specialists. Invoked at the exact phase where each vulnerability would have been caught.

Requirements → threat modelling → code review → IaC → compliance → release gate.

Works in Claude Code, Cursor, Windsurf, Warp, and any MCP-compatible tool.

The 4-Minute Problem

You asked Claude Code to build a file upload feature. It wrote working code in 4 minutes.

It missed:

Every one of these has appeared in real breach post-mortems. AI agents optimise for working code, not secure code. This project embeds the specialists that close that gap — at the exact phase where each issue would have been caught.

What you get

Agents

Who Do You Call?

What are you working on?
│
├── Starting a new feature?
│   ├── product-manager  →  "Define security requirements for X using ASVS L2"
│   └── grc-analyst      →  "Initialise risk register, map to SOC2 / GDPR / PCI-DSS"
│
├── Designing the architecture?
│   ├── appsec-engineer          →  "Threat model this design using STRIDE"
│   ├── cloud-platform-engineer  →  "Review IaC for this feature"
│   └── ai-security-engineer     →  "Security review — feature calls an LLM"  ← always include this
│
├── Writing or merging code?
│   ├── dev-lead       →  "Review PR #N for secure coding issues and dependency risks"
│   └── appsec-engineer  →  "Triage SAST findings for PR #N"
│
├── Quick security question (any phase)?
│   └── security-champion  →  "Is this pattern / library safe? Context: ..."
│
└── Ready to ship?
    └── release-manager  →  "Run pre-release security checklist for vX.Y.Z"

Quick start

Option 0 — Claude Code Plugin Marketplace

/plugin marketplace add Kaademos/secure-sdlc-agents

/plugin install secure-sdlc-agents@secure-sdlc-agents

All 8 agents are immediately available in your session. No cloning, no npm, no file copying.

Option A — Git clone (zero dependencies)

git clone https://github.com/Kaademos/secure-sdlc-agents.git
cp -r secure-sdlc-agents/.claude /your/project/
cp secure-sdlc-agents/CLAUDE.md /your/project/
cp -r secure-sdlc-agents/docs/templates /your/project/docs/

Then use agents directly:

cd /your/project
claude --agent product-manager "Define security requirements for [your feature]"

Option B — CLI tool (recommended for teams)

Published on npm as @kaademos/secure-sdlc. Requires Node.js 18+.

Global install (command is still secure-sdlc):

npm install -g @kaademos/secure-sdlc
secure-sdlc --version
secure-sdlc init

No global install (uses npx; pin a version in CI with @1.0.0):

npx @kaademos/secure-sdlc@latest init

After install — useful commands:

secure-sdlc paths              # print PACKAGE_ROOT and MCP server path (for Cursor MCP JSON)
secure-sdlc init --cursor      # scaffold project + .cursor/mcp.json pointing at bundled MCP
secure-sdlc install-mcp        # merge MCP server into ~/.cursor/mcp.json (and other tools)
secure-sdlc kickoff            # interactive feature wizard
secure-sdlc status

Develop / run from a git clone (no npm publish needed):

cd /path/to/secure-sdlc-agents
npm install
node cli/bin/secure-sdlc.js init
# or: npm run sdlc -- init

Option C — Cursor / Windsurf / Other MCP tools

1. Get the absolute path to mcp/src/server.js:

• If you installed the CLI from npm: run secure-sdlc paths and copy MCP_SERVER.
• If you use a git clone: run npm install at the repo root (installs MCP SDK for the bundled server), then use
  /absolute/path/to/secure-sdlc-agents/mcp/src/server.js.

1. Add to your MCP config:

Cursor (~/.cursor/mcp.json or .cursor/mcp.json in project):

{
  "mcpServers": {
    "secure-sdlc": {
      "command": "node",
      "args": ["/absolute/path/from-secure-sdlc-paths/mcp/src/server.js"]
    }
  }
}

Claude Code:

claude mcp add secure-sdlc -- node /absolute/path/to/secure-sdlc-agents/mcp/src/server.js

Or install for all tools at once:

node cli/bin/secure-sdlc.js install-mcp --tool all

1. Copy the Cursor rules for automatic security context:

cp -r .cursor /your/project/

1. Use the sdlc_* tools in any chat:

Use sdlc_plan_feature to define security requirements for a new payment checkout feature.
Stack is Next.js + Stripe + PostgreSQL. ASVS L2. Compliance: PCI-DSS, SOC2.

The lifecycle — phase by phase

PLAN        product-manager (ASVS requirements)
            + grc-analyst (risk register, compliance mapping)
                    ↓
DESIGN      appsec-engineer (STRIDE threat model)
            + cloud-platform-engineer (IaC review)
            + ai-security-engineer (if AI/LLM features)
            + grc-analyst (compliance gate)
                    ↓
BUILD       dev-lead (PR review, SCA) — on every PR
            + appsec-engineer (SAST triage)
            + cloud-platform-engineer (secrets, pipeline)
            + security-champion (quick questions any time)
                    ↓
TEST        appsec-engineer (DAST, pentest)
            + dev-lead (regression)
            + grc-analyst (audit evidence collection)
                    ↓
RELEASE     release-manager (go/no-go)
            + grc-analyst (compliance attestation)
            + cloud-platform-engineer (production hardening)

Severity gates:

• CRITICAL — blocks all gates, no exceptions
• HIGH — blocks Build→Test and Test→Release without documented accepted risk
• MEDIUM — requires remediation plan or accepted risk before release
• LOW — tracked in risk register, does not block

Frequently Asked Questions

Q: Where do I put my OpenAI or Anthropic API key?
You don't need to provide an API key to secure-sdlc. This project does not make LLM API calls directly. Instead, it acts as an MCP server and prompt-generation engine that feeds specialized security context to your "host" AI tool (Cursor, Windsurf, Claude Code). Your API keys and billing are handled entirely by your host application.

Q: Do I have to manually fill out the Markdown templates?
No. While the project provides structured templates in docs/templates/, you do not fill them out by hand. When you invoke a tool like sdlc_plan_feature, the MCP server passes the blank template to your AI assistant, and the AI automatically writes the completed, project-specific markdown file directly to your docs/ folder.

Q: Do the AI agents run automatically in my CI/CD pipeline?
No, the AI agents are designed to be used locally by developers during the coding process (e.g., in your IDE or terminal). The provided GitHub Actions workflow (secure-sdlc-gate.yml) does not invoke LLMs. Instead, it acts as a deterministic gatekeeper—it runs traditional tools (like Gitleaks, Checkov, CodeQL) and verifies that the AI-generated artifacts actually exist and are fully filled out before allowing a merge.

Q: Will this use a lot of API tokens/credits?
Because this tool feeds comprehensive security frameworks (like OWASP ASVS), infrastructure checklists, and full file templates into your AI's context window, it can consume a significant number of tokens. Ensure your host application (like Claude Code or your Cursor subscription) has sufficient limits for handling large context prompts.

Q: Can I customize the templates for my own company's requirements?
Yes. When you run secure-sdlc init, the default templates are copied into your local docs/templates/ directory. You can modify these markdown files to include your own company's specific compliance headers, and the agents will use your customized versions going forward.

MCP tools reference

When using the MCP server (Cursor, Windsurf, etc.), these tools are available:

CLI commands reference

secure-sdlc init           # Scaffold docs, hooks, CI, config in current project
secure-sdlc init --cursor  # Also install Cursor MCP config and rules
secure-sdlc kickoff        # Interactive wizard to start a new feature
secure-sdlc status         # Show current SDLC phase and artefact status
secure-sdlc review         # Security review a file or diff
secure-sdlc gate v1.2.0    # Run pre-release security gate check
secure-sdlc install-mcp    # Install MCP server for Cursor / Claude Code / Windsurf
secure-sdlc paths          # Show package root + MCP path (after npm install -g)

Git hooks

Included in hooks/:

• pre-commit — secret detection, lock file checks, security anti-pattern scan
• pre-push — artefact gate for protected branches, open finding check

Install:

bash /path/to/secure-sdlc-agents/hooks/install.sh
# OR via CLI:
secure-sdlc init  # installs hooks automatically

GitHub Actions

.github/workflows/secure-sdlc-gate.yml adds:

• Artefact gate — blocks PRs to main/master if required security docs are missing
• Secret scanning (Gitleaks)
• Dependency audit (npm audit, pip-audit)
• IaC scanning (Checkov — Terraform, K8s, Docker)
• SAST (CodeQL — JavaScript/TypeScript, Python)
• Release gate — full pre-release checklist on workflow_dispatch

Copy to your project:

mkdir -p .github/workflows
cp /path/to/secure-sdlc-agents/.github/workflows/secure-sdlc-gate.yml .github/workflows/

Stack profiles

Deep, framework-specific security guidance in stacks/:

Warp terminal workflows

In warp-workflows/ — import into Warp for one-click SDLC automation:

Document templates

docs/templates/ contains pre-formatted templates for every artefact:

Worked examples

Project configuration

Create secure-sdlc.yaml in your project root:

project:
  name: "my-app"
  stack: "Next.js + PostgreSQL"

security:
  asvs_level: L2
  frameworks: [SOC2, GDPR]
  gates:
    build_to_test:
      block_on: [CRITICAL, HIGH]
    test_to_release:
      block_on: [CRITICAL, HIGH]

Generate one automatically: secure-sdlc init

A note on what these agents are — and aren't

These agents produce guidance, not guarantees.

They will help a team ask the right questions earlier, produce consistent artefacts,
and catch common mistakes that would otherwise slip through. They will not replace a
skilled AppSec engineer, a qualified GRC practitioner, or a thorough penetration test.

Every output should be reviewed by a human with relevant expertise before it is acted on
or used as audit evidence. The threat model is a starting point, not a final document.

Security practitioners are right to be sceptical of anything that claims to automate
security away. This project does not make that claim. It makes security practices easier
to start, easier to maintain, and harder to skip — which is most of the battle.

If you find guidance in an agent file that is wrong or dangerously out of date,
please open an issue.

Prerequisites

• Claude Code for sub-agent usage
• Node.js 18+ for the CLI and MCP server
• Optional: npm install -g @kaademos/secure-sdlc for the secure-sdlc command on your PATH
• Any MCP-compatible AI tool for the sdlc_* tools

Contributing

See CONTRIBUTING.md. High-value contributions:

• Additional compliance frameworks (HIPAA, FedRAMP, NIS2)
• Stack profiles for Go (Gin/Echo), .NET, Java Spring Boot
• More worked examples (OAuth flows, payment processing, AI features)
• Integration guides for specific SAST/DAST tools
• Translations of agent prompts

Related

• OWASP ASVS
• OWASP Top 10
• OWASP Top 10 for LLMs 2025
• NIST SSDF
• Model Context Protocol
• Claude Code documentation

Licence

MIT — see LICENSE.