Release Notes

The Waterwheel agent is distributed as a pre-built Docker image, free to use.

Docker image: taojdcn/duotail-waterwheel

Use the latest release

We recommend always running the latest version, 1.4.0.

docker pull taojdcn/duotail-waterwheel:1.4.0

1.4.0 — latest

The current release. It adds test skills — reusable Markdown instructions the agent loads on demand — so a tricky interaction is described once and a repeated flow such as login is shared across tests without copying prose. See the new Create Test Skills guide.

• Test skills — a skill is a folder holding a single SKILL.md (front-matter name and description, then a Markdown body). The agent sees only the name and description while it decides whether a skill applies, and loads the full body only when it does, so a long skill costs nothing until it is used.
• Built-in skills — three ship inside the image under the reserved ww: prefix, each covering an interaction the raw browser tooling handles poorly on its own: ww:clipboard-capture, ww:form-and-dialog, and ww:take-browser-screenshot. The ww: prefix is reserved, so your own skill names can never collide with a shipped one.
• SKILLS_DIR — user skills are read from /agent/skills by default; point SKILLS_DIR at another folder to swap skill implementations per environment without editing a single test. Built-in skills are unaffected by it.
• New commands:
  • load-test-skills — create or, with --force, replace a skill from piped content.
  • display-test-skills — list every built-in and user skill, or print one skill's SKILL.md with --show.
  • delete-test-skills — remove user skills by name (-n) or clear all of them (-a); built-in skills are never touched.
  • delete-builtin-skills — remove a shipped skill by exact name. There is deliberately no -a, and the only way back is recreating the container.
• Agent loops support — the waterwheel-load-test-skills Waterwheel skill loads a project's test skills into the container as part of the pipeline. It requires this release; toolkit v2.0.0 and newer pin agent image 1.4.0.

docker pull taojdcn/duotail-waterwheel:1.4.0

1.3.0

Makes every part of the agent scriptable and introduces the Waterwheel skills for pairing the agent with a code agent in a nested test-and-fix loop.

• Agent loops & skills — a set of Claude Code skills that install, configure, and drive the agent automatically, enabling an autonomous, test-driven web development workflow. See the new Agent Loops guide.
• Non-interactive setup — configure the agent from scripts and CI without the interactive prompts of config-agent:
  • config-ai-provider — apply an AI provider and model directly.
  • set-domain-permission — generate the browser domain allowlist from a comma-delimited list.
  • enable-test-on-host — enable testing against a web app deployed on the host machine.
  • display-ai-config — print the effective provider, model, and token mode as JSON.
• Programmatic test & instruction management:
  • upload-test-task and upload-instruction-file — create or replace markdown tasks and instruction/config files from stdin.
  • reset-test-config — delete test tasks and/or instruction files to return to a clean state.
  • output-context-variables — export context values produced by the latest run as flat JSON for reuse outside the container.

docker pull taojdcn/duotail-waterwheel:1.3.0

1.2.0

Builds on 1.1.0 with richer test configuration and new helper commands.

• preset-context.json — a single file to manage per-run context variables and to configure the test flow (execution order and dependencies between tasks).
• New commands:
  • manage-test-files — add, list, and remove the markdown task files the agent runs.
  • preset-context — manage the context variables and test flow stored in preset-context.json.

docker pull taojdcn/duotail-waterwheel:1.2.0

1.1.0

Added multi-platform support and the first set of command-line tools.

• Multi-platform images — published for both linux/amd64 and linux/arm64.
• New commands:
  • stop-qa — stop the current QA run.
  • check-test-result — check the latest test results.
  • get-failure-detail — print a diagnostic report for the first failed test.
  • config-agent — configure the AI provider, browser permissions, and access to a test website on the host.
  • manage-global-constants — manage global constants shared across all tests.
  • manage-context-variables — manage per-run context variable overrides. (Replaced by preset-context in 1.2.0.)

Documentation for 1.1.0

This site always documents the latest release. To read the documentation that shipped with 1.1.0, check out the 1.1.0 branch of the site repository and build it locally.

1.0.0-mac

The first release, used for beta testing on macOS machines only.

Deprecated

1.0.0-mac is no longer supported and should not be used. Please upgrade to the latest release.