AGGIB
AGGIB is an individual full-stack developer building Stroq, an open-source local action firewall for AI coding agents. Stroq scans untrusted content read by an agent, taints the session when it detects indirect prompt injection, and deterministically blocks dangerous follow-up actions before they leave the machine.
At a Glance
- Individual developers and development teams using AI coding agents
- Teams operating MCP servers and MCP-enabled desktop clients
- Security-conscious software engineering and DevSecOps teams
- Open-source projects and CI pipelines that want to replay agent-attack scenarios
AI Tools by AGGIB
(1)
Stroq
Local Firewall for AI Coding Agents
Discussions
No discussions yet
Be the first to start a discussion about AGGIB
Latest News
Stroq v0.10.1 moved the attack corpus out of the JavaScript bundle and removed the published sourcemap; release notes say there was no behavior change.
Stroq v0.10.0 expanded secret-egress scanning to 2 MiB windows and added the secret.unscannable rule/action class.
Stroq v0.9.0 introduced the MCP stdio proxy for Claude Desktop and other MCP clients without hook APIs.
The release series added the Windsurf adapter (v0.8.0), OpenClaw adapter (v0.7.0), and Copilot CLI adapter (v0.6.0).
Products & Services
Open-source Apache-2.0 engine and policy layer for AI coding agents. It scans tool results, normalizes and matches content against threat rules, taints suspicious sessions, classifies subsequent actions, and applies an ordered deny/ask/allow policy locally.
npm CLI used to initialize adapters, run hooks and the MCP proxy, inspect and verify the hash-chained audit log, clear taint, explain decisions, plant a canary secret, and replay the attack corpus.
A stdio proxy for MCP clients without native hook APIs. It judges tools/call requests before forwarding them and scans results on the way back, with fail-closed behavior for relevant malformed, oversized, or unaddressable calls.
Market Position
Stroq positions itself as a local, deterministic, taint-aware action firewall rather than an agent's ordinary permission prompt, a small regex hook, or a cloud AI-security service. Its differentiators are carrying provenance from read content into the next action, normalization of obfuscated content, a broad gated rule set, tamper-evident auditing, and fail-closed behavior without a network round trip. The alternatives described by its own documentation are native agent permission prompts, ad-hoc regex hooks, and cloud AI-security platforms; the broader agent-security/guardrails category is the relevant competitive set.
Leadership
Founders
Agybay Gibatollayev
Full-stack developer based in Astana, Kazakhstan; the AGGIB GitHub profile identifies him as building Stroq and lists experience/skills across frontend (Next.js, React, TypeScript, Tailwind), backend (Node.js, NestJS, Go, Python), and infrastructure (PostgreSQL, Supabase, Docker, GitHub Actions).
Executive Team
Agybay Gibatollayev
Creator / full-stack developer
AGGIB GitHub profile owner and primary Stroq developer; based in Astana, Kazakhstan, with a stated full-stack development background.
Business Model
Revenue Model
Stroq is distributed as open-source software under Apache-2.0 through GitHub and the public @stroq/cli npm package. The reviewed official sources describe no paid service, subscription, or commercial revenue model.
Pricing Tiers
The CLI is publicly installable from npm and the source is available on GitHub under Apache-2.0; no paid tiers were listed in the reviewed sources.
Target Markets
- Individual developers and development teams using AI coding agents
- Teams operating MCP servers and MCP-enabled desktop clients
- Security-conscious software engineering and DevSecOps teams
- Open-source projects and CI pipelines that want to replay agent-attack scenarios
- Protecting Claude Code, Cursor, Codex, Copilot CLI, Windsurf, and OpenClaw sessions from indirect prompt injection.
- Protecting MCP clients, including clients without hook APIs, from malicious tool results and dangerous tool calls.
- Preventing secret exfiltration through outbound shell commands, web fetches, MCP calls, and external git pushes.
- Blocking encoded or obfuscated execution and suspicious network follow-up actions after an agent reads untrusted content.
- Auditing and replay-testing agent security policies against recorded incidents in local development or CI.