# Showboat

> Create executable Showboat demos that prove behavior.

- Skill: `kevinslin-skills/showboat` (Agent Skill)
- Install (CLI): `npx skillmds@latest add kevinslin-skills/showboat`
- Raw SKILL.md: https://api.skillmd.com/api/skills/kevinslin-skills/showboat/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: kevinslin (https://skillmd.com/u/kevinslin-skills)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/kevinslin-skills/showboat

---


# Showboat

## Overview

Use `uvx showboat` to build Markdown documents that mix short notes, executable commands, and captured output. Reach for it when the deliverable should both explain and prove behavior. This should be used to do live integration tests. You should NOT use showboat to run existing tests. 

## Workflow

1. Confirm the CLI is available with `uvx showboat --help`.
2. Pick the document path. Default to `demo-{num}-{title}.md` in the current workspace unless the user specifies another file. Use the next available integer for `{num}` and a short kebab-case summary for `{title}`.
3. Initialize the document with `uvx showboat init <file> <title>`.
4. Add short context blocks with `uvx showboat note <file> ...`.
5. Record proof steps with `uvx showboat exec <file> <lang> <code>`.
6. Prefer commands that prove the real behavior, such as tests, generated artifacts, or verification commands.
7. If screenshots matter, add them with `uvx showboat image <file> ...`.
8. If a step was wrong or too noisy, remove it with `uvx showboat pop <file>`.
9. Finish by running `uvx showboat verify <file>`.

## Stable Proof Requirements

- Use Showboat to prove functionality works, not just to describe the plan.
- If the raw output is nondeterministic, do not capture it verbatim when the document should later pass `showboat verify`.
- Do write raw output verbatim in a separate file at `demo-{num}-{title}.raw.md` - this is for manual human inspection later
- Common unstable fields to sanitize or summarize:
  - timestamps
  - temp directories
  - thread ids
  - duration lines
  - stochastic model text
- Prefer writing raw artifacts separately, then use `showboat exec` with a deterministic summary command over those artifacts.
- For test output, redact only the unstable fields and preserve the pass/fail substance.

## Good Fits

- Creating a reproducible `demo-{num}-{title}.md` for a feature or CLI workflow
- Capturing a bug reproduction plus the command that fixes or validates it
- Handing off a workflow another agent can replay with `showboat verify`

## Avoid

- Using Showboat when a plain response is enough and no proof artifact is needed
- Capturing secrets, credentials, or private tokens in output blocks
- Leaving known-unstable output unsanitized when later verification matters
- Proving that tests, smoke checks, or integration flows pass