jkelly-dev1

Agentic Review Gate — AI skill for Claude Code

AI community

How to install Agentic Review Gate

This entry records only its repository, not the path inside it, so there is no exact command to give. Open jkelly-dev1/agentic-review-gate and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Agentic Review Gate does

Governed, auditable AI review platform: the review and approval work a team does by hand, as a webhook-triggered LangGraph agent with human-in-the-loop approval, Langfuse tracing, an eval-gated CI pipeline, and a tamper-evident hash-chained audit trail. Provider-agnostic (mock/Anthropic/OpenAI), FastAPI, Bicep IaC for Azure Container Apps.

Alternatives in AI

Axonhub — ⚡️ Open-source AI Gateway — Use any SDK to call 100+ LLMs 3.2k ★
Claude Hub — by Claude Did This - A webhook service that connects Claude Code to GitHub repositories, enabling AI-powered c 393 ★
Seo-agi -- Generative Engine Optimization For AI Agents — You are an elite GEO (Generative Engine Optimization) and Technical SEO agent 171 ★

README

Agentic Review Gate

the review and approval work helps to reduce the manual work.

_By James Kelly._ MIT licensed. Runs fully offline; no API key needed to try it.

A GitHub webhook (PR or issue) triggers a LangGraph agentic workflow that plans a review, retrieves the relevant standards from a small local corpus, drafts an analysis, self-critiques it with an LLM-as-judge, then pauses for human-in-the-loop approval before it posts a result and writes a tamper-evident, fully traceable audit record (prompt versions + model id + inputs + retrieved sources + eval verdict + approver + timestamp).

The pattern applies anywhere an AI decision has to be trustworthy: regulated or safety-critical engineering, finance and compliance, healthcare, legal, content moderation, or any workflow where "an AI approved it" is not good enough and you need to reconstruct exactly what happened and prove the record was not altered. The demo uses code review as a concrete, familiar example; the corpus, prompts, and post-back target are swappable seams. Everything runs fully offline on a deterministic mock provider; the real Anthropic and OpenAI paths are used only when the matching API key is set.

Architecture

   GitHub PR/issue                                          GitHub
   webhook (signed)                                         comment
        |                                                      ^
        v                                                      | post-back
  +----------------------------- FastAPI (app/main.py) -------+--------------+
  |  POST /webhooks/github   HMAC-SHA256 verify + replay guard + rate limit   |
  |  POST /reviews/{id}/approve   human-in-the-loop decision (resume)         |
  |  GET  /reviews/{id}   status + trace     GET /healthz   GET /metrics       |
  +-----------------------------------+--------------------------------------+
                                       | start / resume
                                       v
        +-----