firehose/app/priv/blog/engineering/2026/04-02-automated-release-notes-with-showboat.md
Firehose Bot 2bda8dd58e fix typos
2026-04-09 17:37:46 +02:00

254 lines
8.9 KiB
Markdown

%{
title: "Automated release notes with 'demo' skill",
author: "Willem van den Ende",
tags: ~w(ai),
description: "Agents can make release notes, given the right tools and instructions. Showboat, Chrome DevTools MCP were quite easy to use. ",
published: true
}
---
Yesterday, I added scheduled posts to firehose. I let my coding agent write the [release notes](blog/releases/scheduled-publishing). It is fun to read the release notes like this, and see the screenshots. Do note that a coding agent might not care there is no navigation - it happily will remember the urls and navigate to them, so that is something you might want to double check with exploratory testing :-).
I've included the full skill below. It is not in the repository, it went straight to my home directory so I can use it in other projects. This is the second project I have used it in. It helped me earlier do a fairly large architectural spike for an enterprise client, and include screenshots of a working application in the response document. Working software is still a great measure of progress.
# Demo skill - let your agent generate demos with screenshots for you
---
name: demo
description: >-
Generate a living demo document proving a feature works. Uses showboat
for Markdown assembly with captured command output and rodney for Chrome
browser screenshots. Use when the user says "demo this", "show me it works",
"create a demo", or after /build completes.
argument-hint: "<feature-name> [--scenario <description>] [--plan <path>]"
user-invocable: true
allowed-tools: Read, Write, Glob, Grep, Bash
---
# Demo
Role: worker. This command generates a standalone Markdown demo document that proves a feature works, using showboat for document assembly and rodney for browser automation.
You have been invoked with the `/demo` command.
## Parse Arguments
Arguments: $ARGUMENTS
- Positional: `<feature-name>` (required) — short name or description of the feature to demo
- `--scenario <description>`: Explicit demo scenario describing what to show. If omitted, infer from the plan and recent commits.
- `--plan <path>`: Path to the plan file. If omitted, search `plans/` for the most recently modified `.md` file with status `implemented` or `approved`.
## Steps
### 1. Check tool availability
Verify showboat and rodney are installed:
```bash
showboat --help 2>/dev/null && echo "showboat: ok" || echo "showboat: missing"
rodney --help 2>/dev/null && echo "rodney: ok" || echo "rodney: missing"
```
If either tool is missing, tell the user:
> One or more demo tools are missing. Install them with:
> ```bash
> go install github.com/simonw/showboat@latest
> go install github.com/simonw/rodney@latest
> ```
Do not proceed until both tools are confirmed available.
### 2. Gather feature context
Build an understanding of what to demo from these sources (in priority order):
1. **Explicit scenario** (`--scenario`): If provided, use as primary guide.
2. **Plan file**: Read the plan's Goal, Acceptance Criteria, and completed Steps.
3. **Recent commits**: Run `git log --oneline -15` and `git diff main...HEAD --stat` to identify changed files and commit messages.
4. **Route map**: Cross-reference changed files against known LiveView routes in the router (`lib/hub_web/router.ex`).
From these sources, produce a **demo outline**:
- **Narrative**: 2-3 sentence description of what the feature does
- **Backend evidence**: mix commands, test output, or database queries to run
- **UI pages**: which routes to visit and what to look for
- **Interactions**: any clicks, form fills, or navigation sequences to perform
If no plan exists and commits are ambiguous, ask the user for a brief description of what to demo.
### 3. Check dev server
```bash
curl -s -o /dev/null -w "%{http_code}" http://localhost:4000/ 2>/dev/null
```
If the server is not reachable:
> The Phoenix dev server is not running. Start it now?
> ```bash
> mix phx.server &
> ```
After starting, wait up to 10 seconds and verify connectivity. If it still fails, proceed with backend-only evidence (skip all browser screenshots) and note the limitation in the demo document.
### 4. Handle authentication
The app requires authentication for LiveView routes. Before capturing UI screenshots:
1. Ensure a demo user exists:
```bash
mix run -e "
alias Hub.Accounts
case Accounts.get_user_by_email(\"demo@example.com\") do
nil -> Accounts.register_user(%{email: \"demo@example.com\", password: \"demodemo1234\"})
user -> {:ok, user}
end
"
```
2. Log in via rodney:
```bash
rodney start
rodney open http://localhost:4000/users/log-in
rodney wait "input[name='user[email]']"
rodney input "input[name='user[email]']" "demo@example.com"
rodney input "input[name='user[password]']" "demodemo1234"
rodney click "button[type='submit']"
rodney waitidle
```
If login fails, warn and proceed with backend-only evidence.
### 5. Initialize the demo document
Slugify the feature name (lowercase, hyphens, no special chars). Then:
```bash
showboat init "demos/demo-$(date +%Y%m%d-%H%M%S)-<slug>.md" "Demo: <Feature Name>"
```
Store the demo file path for use in all subsequent steps.
### 6. Narrative introduction
```bash
showboat note <demo-file> "## Feature Overview
<2-3 sentence description derived from the plan or commits.>
**Branch**: $(git branch --show-current)
**Commits**: <N> commits ahead of main
**Plan**: <plan file path or 'none'>
"
```
### 7. Backend evidence
Capture backend proof via `showboat exec`. Always include relevant tests. Add narrative notes between evidence blocks explaining what each proves.
**Test output** (always include):
```bash
showboat note <demo-file> "## Test Suite"
showboat exec <demo-file> bash "mix test <relevant-test-files> --color"
showboat note <demo-file> "All <N> tests pass, confirming <specific criterion>."
```
**Compilation check**:
```bash
showboat note <demo-file> "## Compilation Check"
showboat exec <demo-file> bash "mix compile --warnings-as-errors"
```
**Database state** (if relevant to the feature):
```bash
showboat note <demo-file> "## Database State"
showboat exec <demo-file> bash "mix run -e '<query expression>'"
```
### 8. UI screenshots
For each UI page identified in step 2, navigate with rodney, screenshot, and embed via showboat.
**Static page capture**:
```bash
rodney open http://localhost:4000/<route>
rodney waitidle
rodney screenshot demos/screenshots/<feature>-<page-name>.png
showboat note <demo-file> "### <Page Name>
<What this page shows and why it proves the feature works.>"
showboat image <demo-file> '![<What the screenshot demonstrates>](demos/screenshots/<feature>-<page-name>.png)'
```
**Interactive flow** (form submissions, navigation):
```bash
showboat note <demo-file> "### Interactive Flow: <Flow Name>"
# Before state
rodney screenshot demos/screenshots/<feature>-before.png
showboat image <demo-file> '![Before: <initial state>](demos/screenshots/<feature>-before.png)'
# Perform interaction
rodney click "<selector>"
rodney input "<selector>" "<value>"
rodney click "<submit-selector>"
rodney waitidle
# After state
rodney screenshot demos/screenshots/<feature>-after.png
showboat image <demo-file> '![After: <resulting state>](demos/screenshots/<feature>-after.png)'
```
### 9. Acceptance criteria checklist
If a plan file exists, map each acceptance criterion to evidence:
```bash
showboat note <demo-file> "## Acceptance Criteria Verification
- [x] <Criterion 1> -- see Test Suite output above
- [x] <Criterion 2> -- see <Page Name> screenshot
- [x] <Criterion 3> -- see Database State output
"
```
If no plan, summarize what was demonstrated and what it proves.
### 10. Clean up
```bash
rodney stop 2>/dev/null || true
```
### 11. Report results
Display:
```
## Demo Complete
- **Document**: demos/<filename>.md
- **Screenshots**: <N> captured in demos/screenshots/
- **Evidence**: <N> backend commands, <N> UI screenshots
- **Acceptance criteria**: <N>/<M> demonstrated
```
## Error Handling
- **Tools not installed**: Show `go install` commands. Do not proceed without them.
- **Dev server not running**: Offer to start. If startup fails, produce backend-only demo and note the limitation.
- **Authentication failure**: Proceed with backend-only evidence. Note skipped UI screenshots in the document.
- **Screenshot failure**: Log the error as a note in the demo document, continue with remaining screenshots.
- **No plan found**: Infer from git commits and changed files. Ask the user for a description if commits are ambiguous.
- **Rodney/Chrome crash**: Run `rodney stop` then `rodney start` to reset. Retry once. If it fails again, degrade to backend-only.
## Integration
- `/build` can suggest running `/demo` after a successful build
- `/pr` can invoke `/demo` to generate evidence before or after PR creation
- `/plan` produces the acceptance criteria this command verifies visually
- Beads (`bd`) can reference demo documents as task completion evidence
- Reads plan files from `plans/` directory (same format as `/build`)