> ## Content Index
> Fetch the complete content index at: https://www.tao.media/llms.txt
> Use this file to discover other available public pages before exploring further.

# Ridges Launches Ridgeline for GitHub Coding Tasks and Pull Request Reviews
- URL: https://www.tao.media/ridges-launches-ridgeline-for-github-coding-tasks-and-pull-request-reviews/
- Published: 2026-11-19T18:45:16.000Z
- Updated: 2026-11-19T18:45:17.000Z
- Description: The Bittensor SN62 project brings competition-derived coding agents into a GitHub workflow with human-controlled merges.
- Author: Tristan Hillerich
- Tags: Ridges, Subnets, AI, News

Ridges has [launched Ridgeline](https://x.com/ridges%5Fai/status/2123457539421749455?ref=tao.media), a GitHub-native AI software-engineering service that turns issues into pull requests and reviews code changes inside existing repositories. Developers assign the work, agents prepare the changes, and a human decides whether to merge them. Developers do not need a local IDE.

The product brings coding agents derived from [Ridges.AI](https://www.ridges.ai/?ref=tao.media), the competitive software-engineering network on Bittensor subnet 62, into a developer-facing application. Ridges runs competitions in which contributors submit agents that are evaluated on software tasks and rewarded for improvements in performance or cost.

[Ridgeline](https://app.ridges.ai/?ref=tao.media) offers two connected services: Agents, which implements tasks, and Reviews, which examines pull requests. Both draw from the same credit balance. An implementation task costs one credit, while a review costs 0.1 credits, with access available through purchased credits or locked subnet stake.

## From a GitHub Issue to a Tested Patch

Developers start by installing the [Ridges GitHub App](https://github.com/apps/ridges-ai?ref=tao.media), selecting the repositories it can access, and enabling issues. Opening an issue with the `ridgeline` label dispatches the work. The issue describes the desired result in ordinary language rather than a special prompt format.

The work runs unattended in a Ridges-managed sandbox. Under the documented [agent pipeline](https://docs.ridges.ai/ridgeline?ref=tao.media), the system installs the repository's dependencies and runs its tests before making changes, establishing which tests already pass or fail. Existing failures are outside the task unless the issue specifically asks the agent to fix them.

The agent then converts the issue into a working specification grounded in the repository, plans the implementation, and changes the code. Because it cannot return with follow-up questions, the issue needs to contain the requirements it will need to finish the job. The repository also needs to be installable and testable from a clean checkout.

Verification starts with the narrowest test that covers the change, followed by the broader suite at most once at the end. The target is to preserve every previously passing test. A separate self-review examines the diff for unrelated changes, debug output, scaffolding, broken references, and incomplete requirements before the platform opens a pull request.

Ridges says multiple agents compete on each issue and the strongest patch is selected, rather than returning the first attempt. The product lists Python, TypeScript, Go, Rust, React, Django, Node, and Postgres among the technologies it supports. Developers receive the full diff and can request changes or rerun the task with additional context.

The separation between agent and platform is part of the [security design](https://docs.ridges.ai/ridgeline/privacy-and-security?ref=tao.media). The sandbox receives the task and a repository checkout, but no GitHub credentials or platform secrets. Its output is a patch; the backend applies that patch, pushes a branch, and opens the pull request. Code is processed outside the developer's machine, and portions are sent to third-party model providers for inference.

Merge approval remains with the developer. The app also has permission to change GitHub Actions workflows, however, and Ridges warns that repository configuration can allow those workflows to run with access to CI secrets before pull request review. That makes workflow changes relevant to repository access decisions, not just the final merge.

## Reviews and Fixes Inside the Pull Request

Developers can request a [pull request review](https://docs.ridges.ai/ridgeline/pr-reviews.md?ref=tao.media) by posting `@ridgeline-ai review` on its own line in a comment. Automatic reviews can also be enabled for new pull requests through repository configuration.

Findings appear as GitHub review comments on the relevant lines or code blocks, with an explanation, severity, and a suggested change where the fix is clear. Findings that do not map to the diff appear in the review body. Each review includes up to eight findings, prioritized by severity from critical to low.

A subsequent review examines only commits added since the previous one. Developers can follow up with `@ridgeline-ai fix` to have an agent address the latest review's findings, adding other text to supply instructions or context. That fix is billed as a one-credit task, separately from the 0.1-credit review.

Review charges go to the person requesting the run. The commenter must have a linked Ridgeline account and write access to the repository, preventing outside comments on public pull requests from spending a maintainer's balance.