Gstack is a production-grade AI agent skill framework and headless browser automation toolchain designed for Claude Code, Codex, Cursor, and other coding agents. It provides a structured skill system (50+ slash-command skills like `/ship`, `/review`, `/plan-ceo-review`) with a resolver pipeline that generates per-host `SKILL.md` templates from TypeScript sources. The repo also contains a fully functional headless browser daemon (`browse/`) with anti-bot stealth, multi-layer prompt injection defense (canary tokens, ML classifiers, URL blocklist, envelope sentinel), CDP escape hatch, SOCKS5 proxy bridge, and secure file I/O. Additional subsystems include markdown-to-PDF generation (`make-pdf/`), iOS QA automation (`ios-qa/`), design mockup generation (`design/`), a secret redaction engine, and a cross-session knowledge base (gbrain). The codebase is unusually well-tested with extensive unit, integration, and periodic paid E2E tests against real LLM APIs. While primarily an application, it includes data pipeline components like telemetry aggregation, eval result stores, and a brain cache specification with TTL-based entities and calibration weights, making it a hybrid with strong data engineering discipline.
This repository is a meta-application: a curated, opinionated framework of prompts ('skills') designed to govern the behavior of AI coding agents like Claude Code and GitHub Copilot Codex. It does not contain traditional application code but instead defines a structured software development methodology executed by an AI agent. The system transforms an AI agent from a simple chat interface into a disciplined, stateful development partner that plans, implements, reviews, and maintains code according to explicit workflows. The architecture is a monolith of interdependent markdown files organized into a skill hierarchy. The `ask-matt` skill acts as a router, directing user requests to the appropriate workflow. Core primitives like `grilling` (structured interview) and `domain-modeling` (glossary/ADR creation) are reused by higher-level skills. Execution engines like `implement` and `code-review` drive the actual development cycle. A key innovation is the 'stateful workspace' pattern, where the agent writes decisions to files (GLOSSARY.md, ADRs, specs) to maintain context across sessions and manage token limits. The codebase demonstrates exceptionally high craftsmanship in prompt engineering, with clear patterns for context management, sub-agent delegation, and human-in-the-loop interaction. It is actively maintained with a detailed changelog. However, its complexity is a double-edged sword: the tight coupling between skills and the reliance on specific model behavior creates a fragile ecosystem. The documented open bugs and high cognitive load for users are significant risks, preventing it from achieving an 'exceptional' rating despite its innovative design. The repository is a brilliant but brittle system for professional AI-assisted development.
The agent-skills repository provides a framework for enhancing AI coding agents with structured engineering workflows called 'skills'. Each skill encodes senior engineer best practices across the software development lifecycle (Define → Plan → Build → Verify → Review → Ship) through opinionated, process-driven guidelines. Skills are invoked automatically based on context or via slash commands and integrate with multiple AI agent tools (Claude Code, Gemini CLI, Antigravity, Codex, etc.) through tool-specific adapters. The system includes a meta-skill for dynamic skill routing, reusable agent personas for specialized tasks, and validation infrastructure to ensure consistency and prevent drift. It is designed to prevent agents from skipping critical steps by mandating concrete evidence (tests, build output) rather than relying on subjective assessments.
CodeStable is a framework for AI coding assistants (Claude Code, Codex CLI) that provides a set of 'skills' — thin harnesses that guide AI agents through software development tasks such as feature implementation, bug fixing, refactoring, epic decomposition, code review, and lesson management. The system emphasizes risk-based assurance, evidence-driven decisions, and minimal project memory, with a core philosophy of 'thin harness, thick context'. It is distributed as plugins for both Claude Code and Codex CLI, and includes a meta-skill (eval-cs-skill) for authoring, evaluating, optimizing, and releasing skills. The repository contains extensive experiment infrastructure with 15+ structured experiments covering routing accuracy, end-to-end outcomes, planted-defect recall, and learning transfer across multiple models. It features a sophisticated evaluation pipeline with deterministic oracles, hidden tests, regression tests, and checkpointed resume. The codebase demonstrates high craftsmanship with clean abstractions, strict validation, and cognitive honesty tagging (measured/soft/underpowered). Key architectural patterns include the thin skill pattern (compact SKILL.md with progressive disclosure via references), adapter pattern for AI model execution (Claude Code, Codex CLI, API, mock), deterministic seed building for reproducible experiments, and a risk-to-assurance mapping used generically by all task skills. The project has 8 ADRs documenting design decisions and a v1-to-v2 migration path that reduced 32 skills to 8. While the codebase is well-maintained and actively developed, some experiments are marked underpowered and the learning transfer epic was cancelled without completing primary hypotheses. The evaluation system is complex and tightly coupled to specific model names and macOS sandbox profiles, which may limit portability.