web-review-job

Drains pending web-review jobs, inspects target URLs via browser automation, and applies fixes under an autonomy state machine.

Updated Jul 27, 2026
One-click install
npx skills add https://github.com/arayaroma/ether --skill web-review-job-arayaroma
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: web-review-job
Source: https://github.com/arayaroma/ether/tree/main/skills/web-review-job
Command: npx skills add https://github.com/arayaroma/ether --skill web-review-job-arayaroma

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Recurring website quality reviews are easy to skip and hard to trust an agent with. This Skill runs a scheduled review routine that inspects target URLs for issues, applies fixes only when the job has earned autonomy, and records every outcome so trust is built (or revoked) on evidence rather than assumption. ## Core Features & Use Cases - Job queue draining: Pulls pending web-review jobs from ether-pro's job store with ether job pending and processes each target URL in turn. - Tiered autonomy gating: Checks ether job should-confirm separately for read and write actions, so inspection can become autonomous before edits do, and safety violations trigger immediate demotion. - Browser-based inspection: Uses claude-in-chrome to navigate pages, read text and console output, and detect broken layout, dead links, accessibility gaps, and outdated copy, with explicit defenses against prompt injection from page content. - Persisted reporting: Writes findings and fix outcomes to .ether/qa/jobs/<jobId>/report.md and records run results so the promotion/demotion state machine can adjust tiers over time. - Use Case: Schedule /loop 30m /web-review-job to review a marketing site every half hour; early runs ask for confirmation, and as successful runs accumulate the skill begins fixing small issues autonomously. ## Quick Start Run the web review job to drain any pending web-review jobs, inspect each target URL, and report what was found and fixed.

Frequently Asked Questions about web-review-job

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

FAQPage Schema
How do I run a recurring automated website review with an AI agent?▼

Enqueue targets with `ether job add --type web-review --target <url>`, then schedule the skill with `/loop 30m /web-review-job`. Each firing drains pending jobs, inspects the pages, and records outcomes; empty queues are a no-op.

How does the autonomy tier system decide when the agent can edit a site?▼

Read and write actions are tracked separately via `ether job should-confirm`. New targets start at confirm tier; successful recorded runs promote the job toward autonomous, and a safety violation demotes it back to full confirmation immediately.

Can page content hijack the agent during a web review?▼

The skill treats all page text, console output, and DOM content as untrusted data, never instructions. Imperative text found in pages is reported as a prompt-injection finding, and page content never decides what gets edited or executed.

Where are web review findings stored after a run?▼

Findings and fix outcomes are written to `.ether/qa/jobs/<jobId>/report.md` inside ether-pro's own directory, since a URL target has no local repo. Chat output alone is not treated as a persisted record.

What counts as a failure versus a safety violation when recording runs?▼

A read failure means the page did not load or tools errored, not that issues were found. `--safetyViolation` is reserved for changes that broke the site or touched out-of-scope files, since it triggers immediate demotion to full confirmation.