Verify Gemini CLI Output with Kane CLI

Extending LLM Coding Capabilities to the Browser

With the Kane CLI skill active, Gemini CLI translates validation requirements into plain-English browser checks, runs them in real Chrome, and reads the deterministic outcome.

Note: Give your terminal agent a real browser and a deterministic pass or fail. Start free. Try Kane CLI

Step-By-Step Skill Installation

Gemini CLI uses structured Markdown skill definitions. You can configure the Kane CLI skill globally or scoped to a single repository.

Option A: Global Environment Setup

mkdir -p ~/.gemini/skills/kane-cli
curl -o ~/.gemini/skills/kane-cli/SKILL.md \
  https://raw.githubusercontent.com/LambdaTest/kane-cli/main/.agents/skills/kane-cli/SKILL.md

Option B: Scoped Project Setup

mkdir -p .gemini/skills/kane-cli
curl -o .gemini/skills/kane-cli/SKILL.md \
  https://raw.githubusercontent.com/LambdaTest/kane-cli/main/.agents/skills/kane-cli/SKILL.md

Gemini CLI scans these directories on startup. No extra config is required; it understands how to invoke and parse Kane CLI operations.

The Autonomous Verification Loop

Once the skill is linked, Gemini CLI moves to a continuous closed-loop workflow.

When asked to build or modify a web workflow, the agent applies file updates, constructs a clean kane-cli run --agent command, and runs the check headless. In agent mode it skips human formatting and streams raw NDJSON.

It parses the stream, reads the run_end payload, and handles the outcome programmatically: on a pass it commits, on a fail it uses the structured failure logs and screenshots to fix its own code.

This is the same loop another coding agent closes when Claude Code writes a feature and Kane CLI proves it works.

Skill-Driven Multi-Processing

The skill file instructs the model how to coordinate complex testing. If a task needs multiple disjoint routes tested at once, Gemini CLI uses shell subprocesses to trigger parallel Kane CLI runs. It tracks each background process, parses the distinct logs, and summarizes a scannable report.

Note: Want the agent mode and skill setup behind this loop? Read the Kane CLI docs. Read the docs

The same skill drives interactive, headless, and agent workflows, so once it is installed you can pick whichever fits the moment. See how the three modes of Kane CLI map onto everyday development, or watch a full run where I built a Salesforce clone with Claude and verified it with Kane CLI. In an agentic IDE the same loop is packaged as an installable bundle, which is how the Kane CLI power adds browser verification to Kiro.

For the wider loop, including running Antigravity CLI headless and chaining the browser check onto it in CI, see how to test code written by Antigravity CLI.