Crash triage
The offending code, the blame, whether it is already known, and a filed story naming the suspect commit and a proposed fix.
For the team
Wake up to an alert that’s already triaged.
An alert fires at 02:00, and the first forty minutes go into working out which code it belongs to. With a triage recipe behind the alert, the agent starts on that part while you’re still waking up.
For on-call and incident response
The wait today
Two in the morning is the worst available time to be doing repository archaeology, and it is exactly when the work lands. The stack trace, the blame, the recent merges, and the answer to “have we seen this before” are all sitting there; they just need somebody awake to go and get them.
They ask
Ask in your own words. The answer names its commit.
Mention @PRCHD with the question you’d have saved for a colleague, and it reaches the agent exactly as typed. The agent reads the project’s checkout in Act mode, which changes nothing on disk, and the answer lands in the thread footed with the commit it was read from.
- @PRCHD who else calls this function, and what breaks if it throws?
- @PRCHD this stack trace [paste] — which commit last touched the line it dies on?
- @PRCHD did anything land in the last six hours that touches the queue consumer?
- @PRCHD is this the same crash we saw on the 14th, or a new one?
They run
The jobs that keep coming back, written down once.
Each is a prompt plus a posture that someone on the team writes once. After that, anyone in a private channel the operator enabled for runs starts it with one slash command, and the result comes back in the thread. These are examples for on-call and incident response, not a catalog to install — PRCHD supplies the engine, and each team writes its own.
Regression bisect
The commit a behaviour changed on, from an actual bisect over the real history.
Fix-branch triage
A branch with a candidate fix and a regression test, left waiting for review.
Blast-radius map
Every caller of a function, and what fails if it throws.
Scheduled and triggered
Some of the work is done before anyone asks.
Put the same recipe on a cron line or behind a webhook URL and it starts on its own, so work that used to wait for someone to wake up or find time is already done when people arrive.
- A crash reporter’s webhook, checked for signature and severity and deduplicated on the issue id, firing the triage recipe against the real checkout before anyone opens a laptop.
- A severity threshold in the verifier, so only alerts above it reach a prompt.
- A repeat of an issue already seen, matched on its dedupe key so it is triaged once.
What changes
The alert arrives already triaged, attributed, and written up.
Where it stops. A trigger fires one recipe the operator wrote, under that recipe’s own posture — the delivery fills in the details, and the recipe sets the prompt, mode, and target. A delivery older than three hours expires before it starts, since an alert that old has usually moved on.
Other roles
The same host answers the rest of the team.
Product managers
@PRCHD is the export limit still 500 rows, and where is it enforced?
Product questions get answered from the code, without booking an engineer.
Engineering managers
@PRCHD what’s changed on main since Friday, and what’s risky in it?
The status an engineering manager currently assembles by asking five people assembles itself.
Founders and CTOs
@PRCHD how hard would it be to support a second currency?
Leadership gets ground truth directly, at 22:00, without pulling anyone off anything.
Where your code lives
The computer you own is the computer that runs.
Repositories, worktrees, prompts, transcripts, diffs, command output, attachments, environment values, and the event log stay on the computer you installed PRCHD on — the one that already has your code and your tools. The agent works there, so there is nothing to copy into a hosted environment and nothing uploaded to be indexed.
One small public service handles what a computer with no open port can’t do on its own: it enrolls your computer’s identity, forwards short push notifications, and holds webhook deliveries until your computer collects them. Each delivery is sealed to your computer’s public key the moment it arrives, so the service only ever holds ciphertext, and it is never in the path of your work.
Standing watch
Send this to whoever runs the shared computer.
One person installs it — an afternoon, not a migration. Everyone else asks from a Slack channel they’re already in, and nobody creates an account, installs an app, or pairs a phone.