Skip to content

docs: record the demo again with lox run - #82

Merged
DavidGodefroid merged 2 commits into
mainfrom
docs/76-demo-lox-run
Nov 18, 2026
Merged

DavidGodefroid merged 2 commits into
mainfrom
docs/76-demo-lox-run

Conversation

@DavidGodefroid

Copy link
Copy Markdown
Owner

Closes #76

What changed

  • site/assets/demo.svg is recorded again: the human answers in lox run, not in the lockmq console. Four steps: the broker password typed in lox run, the first connection asked and approved there, a stream read approved by a rule (auto:action in the feed), the messages back with the email masked.
  • scripts/brand/demo.py no longer types the terminals' text: it reads a recording kept in scripts/brand/demo/ (six lox run captures and the agent's calls and results) and only lays it out.
  • README: the "Recorded before decisions moved to lockhub" caption is gone; the alt text lists each step. Same alt text and a new caption on the site (site/index.html, which uses the same file; height 640 → 590).
  • docs/install.md: one line on where the recording lives.
  • Topic security added to the repository.

Where each screen comes from

One real run on 2026-11-18, releases as a user installs them: lockmq v0.10.0 (image ghcr.io/davidgodefroid/lockmq@sha256:08a8cf46…, signature verified by lox add), lox v0.12.1 (lockhub release binary), RabbitMQ 4.4.0 (rabbitmq:4-management) in Docker. Workspace ~/shop, declared with lox add lockmq --name dev --url http://host.docker.internal:15676 --amqp-url amqp://host.docker.internal:5676 --vhost orders --user lockmq-ro --credentials ask, default rules, tier read. Fixture: vhost orders with order-events (stream, 3 JSON messages with a customer email, 2 consumers), orders.retry (classic, 1 consumer), orders.failed (quorum, 1204 messages, no consumer); user lockmq-ro with the monitoring tag.

On screen Source
Right pane, every character and colour tmux capture-pane -e of lox run (106x22) during the run, committed as scripts/brand/demo/frames/*.ansi: 1-password (the console's password question, relayed as a card), 2-password-typed, 3-serving, 4-connect-question (the agent's first request), 5-connected (after y), 6-read-auto (the stream read, auto:action ✓)
Left pane: prompt, lockmq_queues and lockmq_read arguments, the queue rows, the message body with <redacted:r1.1>, the answer scripts/brand/demo/agent.json: the same run's agent side (Claude Code headless, claude -p, with the .mcp.json that lox add wrote), results as the MCP server returned them

What the script adds, and what it cuts (nothing else is written by hand):

  • Added: the two window frames and their titles, the four captions under them, the "you press y" label, the dot that is dim while a call runs and green once it answered, the └ connectors.
  • Left pane layout: each queue is shown as name type messages_ready N consumers N, four fields of the returned item (the full items are in agent.json). The message body is the returned JSON, wrapped after commas; … +2 messages stands for the two other messages; … in the lockmq_read call stands for its intent argument.
  • Cut from the run: the agent's lockmq_status and lockmq_stream calls, the first-start policy review, a lockmq_pii_add call after the read, and the second paragraph of the answer, which reported that call's failure (see below). agent.json names them in calls_not_shown.
  • The lox run frames are not consecutive: two intermediate captures are left out (the connect line before its ✓, the read line before its ✓).

Verified

  • python3 scripts/brand/demo.py > site/assets/demo.svg reproduces the committed file; the SVG parses as XML.
  • Rendered in headless Chromium at t = 2, 7, 11.5, 16 s (below) and with scripts/brand/render.mjs (resvg) at 16 s.
  • sh scripts/hygiene.sh: hygiene: clean.
  • prefers-reduced-motion still pauses the image on a frame where everything shows (16 s).
Before (approval in the lockmq console) before
1. password in lox run 1
2. first connection asked 2
3. stream read on a rule 3
4. masked messages 4

Seen during the recording, not part of this change

  • lockmq_pii_add fails under lox run: pii rules: open <workspace>/.lockmq/pii.toml: read-only file system (v0.10.0), so the mask rule a read proposes cannot be saved by the agent. The email was masked anyway (detectors).
  • lockmq_read with from: "last", count: 3 returns one message (the last), not the last three; an agent asked for "the last 3 messages" needs a second read.
  • The agent skill that lockmq init writes still tells the agent to say "lox approve".
  • The demo is 58 kB (20 kB before).

🤖 Generated with Claude Code

The demo showed the approval in the lockmq console, from before
decisions moved to lockhub. It is now built from a recording of one
real run (lockmq v0.10.0, lox v0.12.1, RabbitMQ 4.4.0): the lox run
terminal captured at six moments, and the agent's calls and results.
scripts/brand/demo.py reads the recording in scripts/brand/demo/ and
types none of the terminals' text itself.

README and site: the 'recorded before lockhub' caption is gone, the
alt text lists the four steps.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
agent.json: the read's intent, recorded in French, is kept in an
English translation and marked as such; run_order and answer_cut say
what the run held and the image leaves out. The alt text attributes
the connect question to the agent's first broker requests. demo.py's
docstring lists what it composes on the left terminal.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
@DavidGodefroid
DavidGodefroid merged commit 8a741f0 into main Nov 18, 2026
7 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Re-record the demo with lox run and add the security topic

1 participant