#llm-security

Security-First Agentic IDE Coding Harness

Eight security skills for Claude Code with no dependencies: standard library only, tested, and usable as a CI step. Secret scanning, Python SAST, prompt-injection testing, plus HTTP, JWT, Dockerfile, CORS and dependency auditing.

Attack taxonomy, toolkit & defenses for persistent-memory poisoning of LLM agents. Author: Krishita Sanjay Choksi.

Runtime permission firewall for AI agents - YAML policies, one-tap approvals (macOS popup/Telegram), tamper-evident signed audit log. Benchmarked: 103/103 attacks blocked, 8/8 normal dev ops allowed.

🐀 Fuzz your verifier before an RL agent does. Static + dynamic LLM security auditor to detect reward-hacking in RL post-training environments (OpenEnv, verifiers-spec, Gymnasium).

Find where agent-controlled intent reaches consequential actions without an authority check. Python + TypeScript + Go. Zero-dependency SAST for AI agents.

Self-host secrets manager with an AI-safety layer. Aliases instead of values; AI agents never see real credentials. Cloud option coming.

EvalBot — local-first chatbot security & quality evaluation (FastAPI + Next.js). Evaluate chatbot answers against your own docs & guidelines with ML/NLP + AI-judge scoring. Apache-2.0.

A living catalog of LLM security research, updated daily from arXiv

Self-hosted privacy firewall for LLM traffic. Detect, pseudonymize, route, and safely restore sensitive data through an OpenAI-compatible API.

Proxilion is the security layer for the agentic workforce. It turns managed AI agents into governed users by enforcing strict cryptographic boundaries on every API call to SaaS like Google Workspace, Salesforce, or Atlassian.

A self-contained AI security lab demonstrating document poisoning, indirect prompt injection, and data exfiltration in RAG systems. Explores the "helpfulness paradox" across local and frontier LLMs.

An audit gate for AI agent outbound messaging: hold the draft, let a senior agent review it (no private-data leak, prompt-injection resistance), then release. Reference implementation.

Save a secret: a small window opens that Claude cannot see, you paste there, it lands in 1Password. Read it back: any item in your vault loads into Claude with one fingerprint check, names only, never the values. Keys you paste in the chat are deleted before Claude reads them. Mac, Windows, Linux. MIT.

Border Guard — AI-native territory sovereignty & self-evolving security OS. 6-module D-S fusion engine, trajectory detection, anchor detection, fission engine, territory adjudicator. Full design + Python implementation.

AI Guardian Firewall — a local, user-space, agent-agnostic firewall that mediates an autonomous AI agent's actions (files, shell, network, services) with a deterministic policy boundary, a tamper-evident audit log, and a human-in-the-loop approval cockpit. No kernel modules. Apache-2.0.

A carnivorous honeypot for autonomous AI agents. Sweet-looking bait. Sticky ending.

Memory integrity for AI agents — Information-flow control against OWASP ASI06 memory & context poisoning.

Open-source AI Security Gateway that sanitizes secrets, PII, and internal context before prompts reach external LLMs.

Drop-in middleware that detects AI bot traffic. Community-maintained signature database + RFC 9421 Web Bot Auth verification. Node + Python + Go. Apache-2.0.

Claude Code scaffold for building AI platforms securely — agent/LLM security, Kubernetes, SRE, observability, identity, and supply chain, grounded in published framework canon (OWASP, NIST, CIS, SLSA). Data-science lanes included.

Runtime security gateway that stops AI agents from being tricked into leaking data — taint tracking + policy enforcement to contain the lethal trifecta.

Static analysis for AI automation workflows. Find prompt-injection paths, overpowered tools, and write-capable agent jobs before they run.

A disciplined 10-stage autonomous agent pipeline (Reflexion, Tree Search, ReAct) for Claude Code and AI coding agents. Lifts local LLMs to senior engineering quality.

Lab notebooks and the SupportFlow sandbox agent for the Agentic AI Governance for GRC & Cybersecurity course. Review a real AI agent that reads customer data, calls tools, and moves money.

Scanner for agent tool output: flags text that mimics harness messages, overrides instructions, spoofs authority, or hides directives in HTML. Claude Code skill + CLI.

Static analysis for AI agents in GitHub Actions — finds where attacker-controlled text reaches an agent's prompt, and what that agent is allowed to do with it.

ai security cli developer-tools agent firewall policy-as-code python anthropic claude-code

My cyber + AI builder portfolio. Six projects, six months. The thought partner side: Obsidian second brain + AI agent setups (Claude Code, Codex, Grok, Muse, Dot). The security side: detection-as-code, hardened LLM triage, crown jewel triage, AWS IR lab, agent skills. Plain-English walkthroughs.

Policy-as-code admission controller for AI agent skills and MCP tools. SkillBOM, lockfiles, and supply-chain baselines.

Free AI-agent risk self-check, security checklist, and USD 59 launch-readiness mini-audit offer.

Source-backed archive of agent prompt-injection incidents, research records, trust-boundary patterns, schemas, and sanitized defensive summaries.

Curated research on AI agent social engineering, manipulated delegation, prompt injection, tool-use boundaries, and agent security evaluation.

A reproducible definition separating agent-directed manipulation from legitimate machine-readable self-presentation. Two mechanically decidable axes, 13 conformance cases.

Dark Promptery: 324 LLM prompt-injection techniques, crosswalked to OWASP/MITRE ATLAS/NIST/CWE.

Multi-tenant agentic RAG lab: four retrieval authz flaws, five defense layers, measured attack success rates

Interactive, OWASP LLM Top 10-grounded teaching lab for LLM/RAG threat modeling. Includes a real-time incident scenario chaining data poisoning, indirect prompt injection, and cross-tenant data leakage.

AISRF, the AI Security & Research Framework: every LLM request from your apps and agents becomes a ticket a human approves before it reaches the backend. Analyzers, guardrails (Rebuff, LLM Guard, NeMo, Lakera), red teaming (garak, promptfoo, PyRIT), source code review, reports, MCP server.

Check before you paste into ChatGPT, Claude or Gemini. Local-first preflight for text and Office/PDF that inspects on your machine and stops the request in the page. Publishes its own measured miss rate (recall 0.30) instead of claiming coverage. MIT.

A security guard for AI assistants. It checks every tool an AI tries to use, preventing data leaks and unauthorized commands.