shipsmooth

a simple, powerful coding workflow

shipsmooth is an AI assistant plugin that gives you a simple yet powerful workflow for coding. Read on, or watch the demo.

shipsmooth makes two assumptions:

  1. You need to plan, but plans will change during implementation!
  2. When implementing, you want to break the planned work into vertical slices, probably taking on the risky bits first and validating them, and only then focussing on code quality.

The below image shows what a typical coding session would look like. Note especially Steps 4, 5A and 5B. They show the part where a developer might choose to keep updating the plan and task list, depending on how the implementation is going.

The shipsmooth workflow Create a plan, break it down to tasks, optionally assign risk levels, then start (or resume) codegen. Check whether everything looks good: small tweaks loop straight back to codegen, big changes go through updating the plan and tasks (versioned in git) before resuming. When all looks good, mark the plan complete. You can pause and resume at any time. Pause and resume any time 1. Create a plan 2. Break down to tasks (optional) 3. Assign risk levels to tasks 4. Start (or resume) codegen 5A. Everything looks good? 5B. Update plan & tasks (versioned in git) big changes? small tweaks? 6. Mark Plan "Complete" Yes
A typical coding session with shipsmooth (watch the demo | read the skill file)

One Minute Demo

Take a minute and just watch shipsmooth in action. Each step in the workflow gets highlighted when the demo video is playing that bit.

1. You discuss and ask it to make a plan 2. It breaks the plan into tasks (plan version 1) 3. You launch the codegen 4. Oops. Something looks wrong! 5. You pause the codegen 6. You update the plan (Version 3; agent updates tasks) 7. Resume execution 8. Stop and go home 😊 9. Come back the next day!

Read more

Plans? Tasks??

A "plan" is just a structured markdown file that describes the design of a unit of work. See an example below (click to enlarge). A plan file may contain a list of tasks.

Example shipsmooth plan: Plan 13 — Local file-based task tracking
Example plan (click to enlarge)
Tasks section of Plan 13, risk-sorted High to Low
Example tasks section (click to enlarge)

A task is the smallest unit of work. A plan is broken down into tasks. Tasks keep getting added/removed/re-ordered as required, as the plan evolves.

Tasks from Plan 79, stored as XML
Example tasks file (click to enlarge)

How does shipsmooth handle changes in plans and tasks?

Both plans and tasks are stored as text files within the project's git repository (externalized options coming soon). Each version of a plan gets its own git tag (eg: plan-10-v1, plan-10-v2). This makes it available permanently for future reference. See below Version 1 and Version 17 (yep, that was a lot of iterating!) of a plan to migrate from Maven to Gradle.

Plan 71 (Version 1) — Gradle skills trial
Plan 71, Version 1 (click to enlarge)
Plan 71 (Version 17) — Maven to Gradle migration
Plan 71, Version 17 (click to enlarge)

License

Apache 2.0. View on Github

One Minute Demo:

Watch a developer go from Version 1 of a shipsmooth plan to Version 3 over two days, based on new knowledge

(View original)

Try shipsmooth

Claude Code

  1. Inside your Claude Code session, type: 
    /plugin marketplace add bitkentech/claude-plugins
/plugin install shipsmooth@bitkentech
  2. Restart Claude Code, because the plugin needs to run some code at session start
  3. In the new Claude Code session type: 
    /shipsmooth:start

Gemini CLI

  1. Install the plugin (one-time): 
    $ gemini extensions install https://github.com/bitkentech/shipsmooth-gemini
  2. In the new Gemini CLI session type: 
    /shipsmooth:start

Codex

  1. Install the plugin (one-time): 
    $ codex plugin marketplace add bitkentech/codex-plugins
$ codex plugin add shipsmooth@bitkentech
  2. In the new Codex session type: 
    $shipsmooth start

Who's behind this?

shipsmooth is being developed by Pramod Biligiri. A freelance software developer and consultant, currently exploring how to develop software using agents.

Pramod Biligiri
LinkedIn Twitter Email