February 2026 Summaries

RWX has released an official skill for coding agents that provides current documentation, run-definition linting, package information, and validation through real RWX runs, aiming to reduce hallucinations and improve CI/CD workflow creation. The skill can be installed in Claude Code through its plugin marketplace or in tools such as Codex, Cursor, and GitHub Copilot via npx skills, then invoked through slash commands or natural-language requests. It uses a “dumb client” approach in which agents load minimal local context and retrieve detailed, current information through CLI commands for documentation search, syntax linting, language-server support, package browsing, run execution, logs, and artifacts. RWX reports early success with creating and improving run definitions, including migrations from services like GitHub Actions, while emphasizing that deterministic command-line tools provide more reliable feedback than web-fetching documentation. Use of the skill has also exposed documentation ambiguities, such as implicit task dependencies, and supports an evaluation framework that tests headless agents against fixtures and analyzes their errors, inefficiencies, and confusion.

RWX now supports triggering automation runs from third-party webhooks, enabling teams to respond to PagerDuty or Datadog alerts by analyzing incidents, rolling back deployments, or executing custom workflows. Webhook triggers are defined as infrastructure as code in a run definition’s `on` section, with incoming headers and payload data made available to tasks through initialization parameters. The workflow can verify webhook authenticity using provider-specific signature-verification packages, such as PagerDuty’s verification package, before proceeding with automation. After publishing a webhook definition, users can find its endpoint URL and token in the RWX Runs page and configure external providers to send POST requests with the token in the `x-rwx-webhook-token` header. This expands RWX’s existing cron- and manually triggered automation capabilities with event-driven workflows.

AI-assisted development is reshaping the role of version control commits and continuous integration by allowing validation to occur before code is pushed, rather than requiring engineers to use commits primarily to trigger remote tests. For large projects with slow local test suites, this shift depends on CI workflows that can be launched directly from a command-line interface without a Git push, along with content-based caching that avoids rerunning jobs whose inputs have already been tested. Under this model, an AI agent can run CI synchronously, address failures, and then push a commit that is already expected to pass immediately. RWX’s v3 CLI supports this workflow through commands such as `rwx run .rwx/ci.yml --wait`, enabling teams, including prompt engineering groups, to verify CI status before committing and pushing changes.