Compare · CI automation
For the team
What runs when the question isn’t a script?
CI answers the same question on every commit, the same way, and it should. The questions with no script behind them wait for a person with time to look. PRCHD runs the recipe your team wrote for them, on a computer the team owns.
CI
PRCHD
A verdict on a commit, or an answer to a question.
| CI automation | PRCHD | |
|---|---|---|
| What starts it | A commit, a tag, a schedule | A mention, a slash command, a webhook, or a cron line |
| What it runs | Steps you wrote, in order | A recipe — a stored prompt with a fixed posture |
| The same input twice | The same result | The same shape of answer; the wording can vary |
| What it is for | Gates, checks, builds, releases | Judgement work on a question |
| Where it runs | A runner your provider or your team hosts | The computer the team installed PRCHD on |
| What comes back | A pass, a fail, and logs | A thread, a filed ticket, or a branch to review |
Where CI is better
Six reasons your pipeline stays.
CI does each of these better than a recipe would. PRCHD sits beside it and takes the questions that need judgement.
- It is deterministic
The same commit gives the same result today and in six months. A prompt makes no such promise, and shouldn’t be asked to.
- It can block a merge
A red check is authority. An answer from an agent is an argument, and it still needs a person at the review and at the merge.
- It runs on everything
Every push, every branch, every pull request, as many at once as the runners allow. A recipe fires on a question or an event, rather than on every commit.
- It is light enough to repeat
A scripted check can run a thousand times a day. Work that reads code with a model is heavier, so it is aimed at the questions that were waiting for a person.
- Nobody has to read a green build
A pass needs no reading. Everything a recipe produces is written to be read, and reading it is the job.
- It already exists and is trusted
Your pipeline works, and your team believes it. Every check in it stays exactly where it is.
What judgement work looks like
Write the question once. It runs whenever it comes up.
Each row is a recipe a team writes for itself — a stored prompt with a posture. PRCHD runs it and supplies four ways to fire it; the questions are yours.
| The question | What a team writes | What comes back |
|---|---|---|
| Is this advisory reachable in our code? | An Act-mode recipe, fired by the advisory webhook | The call paths that reach it, or a note that nothing does |
| Does this ticket describe behaviour we actually have? | An Act-mode recipe, started from the channel | A comment on the ticket naming the code that already does it |
| What could have caused this metric to move? | A scheduled recipe reading the warehouse and the deploy history | The changes inside the window, ranked, posted in the channel |
| Why did this build go red? | A failure-analysis recipe on the CI webhook | The likely cause and the suspect commit, in the build channel |
They compose
Different tools, pointed at the same repository.
- A build goes red; the CI webhook starts a failure-analysis recipe, and the pipeline stays exactly as it was
- A Work-mode recipe leaves a branch, and CI runs on that branch like any other
- An advisory arrives; the recipe answers reachability, and CI still enforces the version bump
- A recipe posts what it found; the merge stays a person’s decision
- A trigger runs the stored recipe the operator wrote, under that recipe’s own posture; the prompt, mode, and target are set before any delivery arrives
Where PRCHD sits
CI proves the change. PRCHD answers the question that came before it.
CI fires on a commit and returns a verdict. PRCHD fires on a mention, a slash command, a webhook, or a cron line, reads the working tree on a computer the team owns, and returns something a person reads. Keep CI. PRCHD is for the work that used to sit in a queue until someone had an afternoon.
One recipe. Four ways to fire it.
The pipeline stays where it is. The questions around it get a recipe of their own.