securityagent-core 3.1.0

Local-first DLP engine and security layer for AI coding agents

securityagent-core

Shared DLP engine, security skills, and policy enforcement for AI coding agents. Used by SecurityAgent and AgnosticSecurity.

What's inside

Package	What it does
endpoint_agent/	DLP engine — file blocking (60+ path patterns), confidence-scored content scanning (PII, credentials, semantic disclosure detection) with structural validators (Luhn, SSA rules, entropy) and context-aware scoring, data flow taint tracking (tags sensitive data at ingress, detects at egress via hash/n-gram/substring matching), tool call argument scanning (MCP/function call DLP + taint registry), behavioral monitoring (8-signal anomaly scoring), honeypot traps, credential scanning. LLM intent analysis with llama3.1:8b default + fallback chain. Notebook-aware .ipynb extraction. PDF extraction via 3-tier pipeline (PyMuPDF → Tesseract OCR → pypdf fallback) with encrypted PDF detection
skills/	Agent-agnostic MCP skills — secure_read, secure_exec, analyze_prompt, scan_output, check_policy, get_session_policy, audit_log
policy/	Policy engine — per-session least privilege, behavioral chain detection (11 attack patterns), audit trail with agent attribution. Obsidian vault integration via memory_bridge.py for cross-session policy context
obsidianMemory/	Obsidian vault — daily session logs, second-brain knowledge base, used by memory_bridge.py for persistent threat/policy tracking
plugin.py	Standalone CLI entry point + validate_exec(), validate_prompt(), validate_output() pure functions

Install

pip install git+https://github.com/kaushikdharamshi/securityagent-core.git

# With optional dependencies
pip install "securityagent-core[cloud] @ git+https://github.com/kaushikdharamshi/securityagent-core.git"
pip install "securityagent-core[llm] @ git+https://github.com/kaushikdharamshi/securityagent-core.git"

Usage

Python SDK

from skills.adapters.python_sdk import SecurityAgentSDK

sdk = SecurityAgentSDK(agent_id="my-agent", agent_type="langchain")

# DLP-gated file read
result = sdk.secure_read("/path/to/file")

# Command validation
result = sdk.secure_exec("ls -la")

# Prompt intent analysis
result = sdk.analyze_prompt("Get me all customer SSNs")

MCP Server

python -m skills.adapters.mcp_server  # after pip install

Any MCP-compatible client (Claude Code, Copilot, custom agents) can connect via stdio JSON-RPC.

CLI

python -m plugin ~/.env                                    # File read gate
python -m plugin --exec "cat ~/.env"                       # Exec validation
python -m plugin --prompt "Get all passwords"              # Prompt analysis
python -m plugin --skill secure_read --params '{"path":"~/.env"}'  # Skills layer
python -m plugin --mcp-server                              # MCP server

Claude Code Integration

See integrations/claude_code/ for PreToolUse hook configuration.

Tests

pip install -e ".[dev]"
pytest

License

MIT