The Claude Agentic Framework enforces a Spec → Build → Review pipeline with 20 specialized agents and PowerShell hooks, preventing Claude Code from coding too early or finishing incomplete.
What Changed — A configuration layer that constrains Claude Code to a disciplined pipeline
Claude Code is powerful, but it has a common failure mode: it starts coding before requirements are clear, skips edge cases, and sometimes declares work done when it's incomplete. Developer Tomas Rampas felt this pain and built the Claude Agentic Framework — an open-source configuration layer that sits on top of Claude Code CLI and enforces a strict Spec → Build → Review pipeline.
This is not a new AI tool. It's a collection of CLAUDE.md rules, specialized agent definitions, PowerShell hooks, and MCP integrations that constrain Claude Code to follow a repeatable process. The framework currently includes around 20 specialized agents covering roles like product-owner, system-architect, security-specialist, and language experts.
What It Means For You — Concrete impact on daily Claude Code usage
If you've ever had Claude Code jump straight into implementation without understanding the full scope, or finish a session leaving half the requirements unmet, this framework directly addresses those pain points. The core pipeline works like this:
-
/delegatestarts the process. If no specification exists, it automatically triggers/spec. - Spec phase: The agent asks questions one by one, scores its own understanding, and only proceeds when you approve the status.
- Build phase: The system maps every requirement to concrete files before implementing.
-
Review & Fix Loop: Runs up to three times. Three reviewers check the work:
-
spec-compliance-reviewer— checks every requirement one by one -
code-review-gatekeeper— looks at code quality before commit -
peer-review-critic— final independent review (last gate)
-
- Enforced Stop Gate: Two PowerShell hooks prevent the agent from ending the session if peer review isn't satisfied.
The framework also includes a Self-Scoring Loop for improving non-code output: Rubric → Score → Name Weak points → Rewrite → Rescore, continuing until improvement is minimal.
Try It Now — How to install and use the framework
- Clone the repository:
git clone https://github.com/tomas-rampas/claude-agentic-framework
cd claude-agentic-framework
Follow the installation instructions in the repo. You'll need Claude Code CLI installed already.
Start a session with
/delegate:
claude code
# Then type: /delegate
The framework will check for an existing specification. If none exists, it runs /spec automatically. You'll be guided through the Spec → Build → Review pipeline.
Pro tip: The framework includes an anti-drift system that keeps agent registry, documentation, and files consistent. Use it for any non-trivial feature or refactor where you want guarantees about completeness.
Caveat: Quality still depends on how good your individual agent definitions are. The framework provides structure, not magic. But for teams tired of "hope the agent does the right thing," this is a significant upgrade.
Source: dev.to
[Updated 22 Jul via devto_claudecode]
Meanwhile, OpenAI has shipped a first-party plugin—codex-plugin-cc (Apache-2.0, v1.0.6 as of July 8, 2026)—that runs Codex inside Claude Code CLI [per dev.to]. The full install requires four slash commands, not the two widely advertised: /plugin marketplace add openai/codex-plugin-cc, /plugin install codex@openai-codex, /reload-plugins, then /codex:setup. Usage is opt-in for a review gate that can create a Claude/Codex loop, and billing remains separate between the two models. The plugin reuses local Codex auth and works with a ChatGPT subscription (free tier qualifies) or an OpenAI API key.
Originally published on gentic.news
Top comments (0)