Walkthrough: a site and a camera fix
Follow two real runs: a landing page built and put live on a web server, and a camera fault fixed because a document explained it.
On this page
This page follows two runs, quoted from the threads as they happened. They
took place on a demonstration setup: web-01 and cam-01 are example
machines behind a gateway, and the camera fault was set up on purpose.
Ysra's own words are quoted as it wrote them. The cards you approve are described rather than reproduced; Deploy a folder and Write a file show what they contain.
1. A landing page, live on web-01#
The request#
The whole brief was one message, in a thread on the gateway's folder:
Create a new page: a simple landing page for "Ysra Ops", with a headline, three
features (fleet runs, signed approvals, investigate and fix) and a contact email
ops@example.com. Then serve it on web-01.
Ysra didn't need to ask anything: the message said what to build and where to serve it. When either is missing, it asks.
Ysra looks before it changes anything#
I’ll confirm the completed landing-page files and inspect web-01’s active nginx document root, then I’ll deploy the site into its own directory and point nginx at it.
web-01 is healthy but still uses Debian’s default /var/www/html site. I’ll preserve that system site, deploy Ysra Ops to /var/www/sites/ysra-ops, and replace only the active server block with a minimal static-site configuration.
It chose a folder of the site's own and left the system's default site alone.
The files go over, with the old version kept#
The active file confirms there are no other routes to preserve. I’m deploying the five static assets now; the deployment tool will retain a rollback copy of any previous site directory.
The deployment contains exactly the five intended site files.
Before anything was sent, a deploy card listed every file, what was left out on purpose, and where the previous version would be kept. See Deploy a folder to a machine.
The web server's configuration#
The active path is a symlink, so the safe write was refused without changing anything. I’ll update its real source file at /etc/nginx/sites-available/default instead.
Ysra never writes through a symbolic link. It found the real file and wrote that one, after a card showing the diff.
The configuration is written and the previous version is backed up. I’ll run nginx’s syntax test now; only a successful test will be followed by a reload.
nginx accepted the new configuration. I’ll reload it and then query the live homepage and public metadata files directly on web-01.
The syntax test is a read; the reload is a service change. One signed approval covered both, as described in Deploy a site to your server.
Checking that it really answers#
The live homepage now contains the correct headline, all three requested features, metadata, and ops@example.com. I’ll finish by checking the stylesheet, crawl files, and nginx service state so the deployed page is complete rather than HTML-only.
All deployed assets return successfully and nginx remains active; the stylesheet is served as CSS, while robots and sitemap point to web-01.
The closing message#
Deployed the five static files to /var/www/sites/ysra-ops, updated nginx’s default server to use that directory, validated the nginx configuration, and reloaded it. Live verification on web-01 returned the Ysra Ops HTML, HTTP 200 CSS with text/css, valid robots and sitemap responses, and an active nginx service. The previous nginx configuration is backed up at /etc/nginx/sites-available/default.ysra-bak.20261007T013203.
It also said what it could not check:
The running product could not be rendered in this review, so its visual result is unverified.
The gateway has no browser, so the page's look wasn't checked; everything that could be checked from the machine was. Ysra says so rather than claiming more.
2. A camera that wouldn't start#
The runbook Restart camera has one step: restart the camera service, then
check it is running. A signed approval allowed service, config and packages
changes on cam-01, and never delete or power:
You approved: service, config, packages on cam-01, never delete, power, until 2026-10-07 16:01 UTC. Changes outside this still need your OK.
First run: needs a person#
The check failed, and Ysra investigated. It found the cause, but nothing it had read said how to fix it, so it changed nothing:
cam-01 failed at step 1 and needs a person. What I found: The camera service is inactive because it rejects config version 2 and expects version 1. Restart has already failed, and the runbook provides no command-based recovery; it explicitly says to open a case with the camera vendor. No service journal entries are available.
Run of “Restart camera” (version 1) finished: 1 need a person. Didn’t finish: cam-01 at step 1.
What changed: a document#
Someone added the team's known-issues.md to the workspace's files. One
entry, KI-3, described exactly this fault and its fix.
Second run: fixed, with the source named#
Same runbook, same fault:
cam-01 failed at step 1. I looked: The camera service is inactive because its configured version is unsupported. Service status reports "camera: unsupported config version '2'; expected 1", and /etc/camera/camera.conf contains version=2. The retrieved known issue KI-3 directly covers this cause and prescribes setting version=1 and restarting camera. Following “known-issues.md”, I ran sed -i s/^version=.*/version=1/ /etc/camera/camera.conf; systemctl restart camera and the check passes now.
The machine showed Fixed.
The only difference between the two runs was the document. Ysra takes a fix only from the runbook step's If it fails or from a passage its search actually returned, and names the source. Without a source, it stops and says what it found.
Keeping the fix#
Ysra then offered to put the fix where it belongs. In the current wording:
On cam-01 I fixed step 1 of “Restart camera” by following “known-issues.md”: sed -i s/^version=.*/version=1/ /etc/camera/camera.conf; systemctl restart camera. Want me to add that to step 1’s “If it fails”, so it’s in the runbook next time?
Accepting makes a new version of the runbook, so the next run fixes the fault without needing the document.
Where to go next#
-
The same journey as a guide you can follow.
-
Signing ahead, and how investigate-and-fix works.
-
Steps, checks and If it fails.
Related#
- How a Code thread works — the states these runs went through.
- Publish or Deploy — putting a site live.
- Machines — running steps on your own servers.