agent-review-loop

Runs a bounded pull-request review state machine with interchangeable Antigravity and Codex operators.

Updated Jul 19, 2026
One-click install
npx skills add https://github.com/DagimAlemayehuu/AgenticEngineering --skill agent-review-loop-dagimalemayehuu
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: agent-review-loop
Source: https://github.com/DagimAlemayehuu/AgenticEngineering/tree/main/agentic-engineering/core/dependencies/orchestration/agent-review-loop
Command: npx skills add https://github.com/DagimAlemayehuu/AgenticEngineering --skill agent-review-loop-dagimalemayehuu

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve? Supervising agent-authored pull requests safely requires structured evidence, deduplication, and hard limits on autonomous action. This Skill provides an auditable review loop that treats Antigravity and Codex as interchangeable review operators while keeping humans in control of scope approval and final acceptance. ## Core Features & Use Cases - Provider-neutral operator selection: Selects the first available operator from a policy-defined order, so a run never depends on one named product. - Bounded correction loop: Permits exactly one evidence-based Jules correction per run, then blocks for human triage instead of looping autonomously. - Safety-first intake: Validates signed GitHub PR events, allowlists repositories, deduplicates by head SHA, and marks fork PRs as static-only. - Use Case: A team receives a Jules-authored PR; the loop intakes the event, assigns an available Codex operator, evaluates the structured review result, requests one bounded correction on failure, and presents evidence for human acceptance on pass. ## Quick Start Use $agent-review-loop to supervise this pull request with an available Antigravity or Codex operator.

Frequently Asked Questions about agent-review-loop

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I automate pull request review with AI agents safely?▼

Use a bounded review state machine that intakes signed GitHub PR events, selects an approved operator, and evaluates structured review results. Human gates remain at scope approval and final acceptance, and the controller never merges, deploys, or changes repository settings.

How does the review loop choose between Antigravity and Codex?▼

The controller reads operatorOrder from the review policy and selects the first operator whose availability flag is true. Both operators have identical authority and return the same review-result schema, so either can complete a run.

Can the review loop handle fork pull requests?▼

Yes, but fork PRs are marked static_only. The controller does not execute their code, restore their artifacts into a privileged runner, or expose secrets to them.

What happens when an AI-authored pull request fails review?▼

For a Jules-authored PR, the controller emits exactly one jules_correction_requested action with bounded findings. On the next failure, or any failure from a non-Jules PR, it blocks and creates a human-visible remediation record.

Why does a new commit SHA restart the review run?▼

A new head SHA invalidates prior evidence, so the controller treats it as a new run and never reuses a previous review approval. Runs are deduplicated by repository, PR number, and head SHA.