cag
v0.1.1Codex Agent Graphs, natively packaged for Claude Code: project-start, task-delivery, research, continuous-improvement, development-recovery and agent-graph-builder skills sharing one canonical source tree with the Codex install. Stage 3 adds generated subagents.
By ArtemArziLicense: MIT0 GitHub starsUpdated 1 hour ago
Directory evidence
- Runtimes
- Claude Code
- Parsed components
- 10 skill or MCP entries
- Source updated
- Sep 29, 2026
- Manifest status
- Canonical path parsed
The directory validates manifest shape and source location. It does not execute the plugin or provide a security endorsement. Review the indexing methodology →
Install cag for Claude Code
claude plugin marketplace add IchenDEV/agent-plugin-mkt
claude plugin marketplace update agent-plugin-marketplace
claude plugin install cag@agent-plugin-marketplacePaste and run these commands in a terminal with Claude Code. They add and refresh the PluginsMP catalog, then install this plugin.
The installer fetches third-party code from the source repository shown on this page. This directory validates manifest structure and source location, but does not perform a security audit; review the manifest, components, and source before installing.
Get the source manually
git clone https://github.com/ArtemArzi/codex-agent-graphsClone the source repository, then follow its setup instructions to add the plugin to a compatible client. The repository root is the plugin root.
Plugin files
├── .claude-plugin/plugin.json├── skills/agent-graph-builder/SKILL.md├── skills/ai-regression-testing/SKILL.md├── skills/continuous-improvement/SKILL.md├── skills/dev-policies/SKILL.md├── skills/development-recovery/SKILL.md├── skills/production-audit/SKILL.md├── skills/project-start/SKILL.md├── skills/research/SKILL.md├── skills/task-delivery/SKILL.md└── skills/verification-loop/SKILL.md
Included Skills10
Create, refactor, or standardize Codex-native agent graph skills with a shared model-first contract, deterministic state controller, bounded optional agents, receipts, compatibility, and adversarial tests. Use when a user asks to create an agent graph, turn a repeated workflow into a graph-backed skill, align several graph skills to one contract, or audit whether a proposed graph is too complex. Always layer this skill on the system $skill-creator rather than recreating generic skill scaffolding.
Choose and strengthen behavioral regression tests for AI-assisted code changes and bug fixes, including new critical rules before their first incident. Use for test-impact analysis, regressions and mock/real-path gaps. Respect project requirements; do not replace task orchestration or nondeterministic agent evals.
Найти в репозитории одну доказуемую проблему и безопасно довести её до no-op, issue-ready или проверенного commit через существующий Task Delivery. Использовать, когда пользователь просит автономно пробежаться по проекту, найти и исправить точечный баг, улучшать код по расписанию, разобрать падающий CI или явно вызывает $continuous-improvement. Граф ограничивает поиск одним кандидатом, запрещает высокорисковую автономную работу и не заменяет Task Delivery.
Load the repository-wide development policies (development recovery, large-codebase discovery) on demand. Use when planning recovery from spec/plan/implementation divergence or when exploring a large codebase, and whenever a graph skill references the global policies. Under Codex these policies are injected into ~/.codex/AGENTS.md by install.py; under the Claude Code plugin this skill is the delivery path.
Recover implementation work when code, tests, scope, or observed behavior diverges from an accepted specification or implementation plan. Use after a direct contract contradiction, repeated attempts without new evidence, an unplanned workaround or material scope increase, pressure to weaken a test or specification to fit the code, or suspicion that an earlier checkpoint rests on a false assumption. Classify the first false assumption, then choose a safe rebuild from a verified checkpoint or a repair forward while preserving Git evidence. Apply independently of Project Start, Research, and Task Delivery.
Local-evidence production readiness audit for shipped apps, pre-launch reviews, post-merge checks, and "what breaks in prod?" questions without sending repo data to an external audit service.
Быстро подготовить новый или существующий репозиторий к разработке и затем поддерживать его каноническую документацию. Использовать для основания проекта, восстановления карты документации, синхронизации документов после изменений, создания и обновления корневых или вложенных AGENTS.md и явного вызова $project-start. Не использовать вместо реализации одной программной задачи: после готового основания передать её task-delivery.
Run fast, adaptive, evidence-backed research with native Codex tools, installed skills, MCP/apps, optional durable tracking, and bounded multi-agent depth. Use when the user asks to research, investigate, compare, verify current facts, survey a field, or produce a cited decision-ready answer; use a skill-only root loop by default and admit the graph only for deep, resumable, persistent, or independently verified work.
Провести одну программную задачу от исследования и плана до реализации, тестов и проверки результата через быстрый skill-only, отслеживаемый или независимо проверяемый маршрут. Использовать для небольшой локальной правки, режима только плана, реализации по уже проверенному плану, полного plan-to-code цикла, исправления или явного вызова $task-delivery. Root владеет решениями и интеграцией; при полезной изоляции bounded auxiliary agents могут выполнять независимые discovery, preparation, implementation, synthesis или focused review. Controller подключается только для resumability, durable scope/baseline, handoff, сохранения packet/receipt делегированных частей или повышенного риска. Обычные локальные этапы сами по себе не требуют controller.
Select and run proportionate checks for a software change, preserving real command outcomes and reporting evidence and gaps. Use to verify an implementation, refactor, bug fix, or release candidate. Follow project-owned quality requirements; do not invent a coverage quota or replace task orchestration.
Plugin manifests1
{
"name": "cag",
"version": "0.1.1",
"description": "Codex Agent Graphs, natively packaged for Claude Code: project-start, task-delivery, research, continuous-improvement, development-recovery and agent-graph-builder skills sharing one canonical source tree with the Codex install. Stage 3 adds generated subagents.",
"author": {
"name": "ArtemArzi"
},
"repository": "https://github.com/ArtemArzi/codex-agent-graphs",
"license": "MIT",
"keywords": [
"agent-graphs",
"workflow",
"project-start",
"task-delivery",
"research",
"codex",
"dual-harness"
],
"skills": [
"./skills/"
]
}For maintainers
If you maintain this plugin, link to this source-backed listing from your README so users can review its manifest and indexed components.
[cag on Agent Plugins Marketplace](https://pluginsmp.com/plugins/cag)