HAIEC Agent Security

Security checks for AI coding agents before code reaches production.

Status

PRE-RELEASE — UNDER ACTIVE DEVELOPMENT

This project is in early development. No security scanning capability is functional yet. Do not rely on this repository for production security checks.

Related MCP server: corbat

What This Project Is

HAIEC Agent Security is a control surface for AI agent security. It defines four independent security checks that an AI coding agent can invoke when the situation calls for them:

One Workflow, Four Independent Checks

SCAN AI CODE          → scan_ai_security
CHECK TENANT BOUNDARIES → scan_tenant_isolation
CHECK MODEL INTERACTIONS → verify_llm_content
GATE THE RELEASE      → check_deploy_security

These checks are independent. HAIEC does not automatically cascade every engine. Each check is selected on its own semantic merit:

• Source-code security → scan_ai_security

• Cross-tenant boundaries → scan_tenant_isolation

• Actual LLM input/output → verify_llm_content

• Merge/release/deploy → check_deploy_security

Architecture

haiec-ai-agent-security-free-mcp/
├── src/
│   ├── contracts/          # Shared types: findings, results, tools, verdicts
│   ├── mcp/                # MCP server factory and tool definitions
│   ├── security/           # Output sanitization, path boundaries, secret redaction
│   ├── provenance/         # Canonicalization and digest utilities
│   ├── engines/
│   │   ├── ai-security/        # AI source-code security engine (not yet integrated)
│   │   ├── tenant-isolation/   # Tenant isolation engine (not yet integrated)
│   │   └── llmverify/          # LLM content verification engine (not yet integrated)
│   └── orchestration/
│       └── deploy-security/    # Release gate orchestration (not yet integrated)
├── tests/                  # Unit and architecture tests
├── evals/                  # AI tool-selection evaluations
└── rules/                  # Rulepack metadata (no rule bodies published yet)

Engine Independence

Each engine is an independent product with its own execution path:

• scan_ai_security does NOT invoke LLMVerify or Tenant Isolation.

• scan_tenant_isolation does NOT invoke LLMVerify or AI security scanning.

• verify_llm_content does NOT invoke source scanning or tenant scanning.

• Only check_deploy_security may intentionally compose engines, and it must disclose which engines ran, which did not, and why.

Current Capabilities

What exists today:

• TypeScript package scaffolding with MCP v2 structure

• Shared contracts (findings, results, tools, verdicts)

• Tool metadata with semantic "when to use" / "when not to use" descriptions

• Engine independence enforcement (architectural tests)

• Path boundary validation

• Secret redaction in outputs

• Output sanitization

• Canonicalization and provenance helpers

• AI tool-selection evaluations

• CI workflow

What does NOT exist yet:

• No security scanning engine is integrated

• No MCP tool is registered for execution

• No rule bodies are published

• No npm package is published

• No security guarantee is provided

Installation

This package is not yet published to npm. It is intended for local development only at this stage.

git clone https://github.com/subodhkc/haiec-ai-agent-security-free-mcp.git
cd haiec-ai-agent-security-free-mcp
npm ci
npm run build
npm test

License

License decision pending. Until a license is added, this repository should not be described as open source. The final license remains a founder/IP decision.

Contributing

This project is in pre-release development. Contributions are not yet solicited until the foundation stabilizes and a license is selected.

Related Projects

• LLMVerify — LLM content verification (independent product, separate repository)