@whittlelabs/sifter 0.30.9 → 0.30.11

package/README.md

@@ -82,6 +82,7 @@ The Sifter subscribes to your job pool and waits for review jobs. Leave it runni
 A running Sifter deals with Sift or `claude` going away on its own, and says what it is doing:
 
 - **Sift can't be reached** (a deploy, a network blip): it logs one warning, keeps trying, and says when Sift is back. This holds at startup too, so starting it during an outage is fine; it begins claiming once Sift answers.
+- **Sift asks it to slow down** (too many sign-in attempts from this machine in a short time, which Sift answers by locking it out for a while): this is not a problem with the pairing, so it logs one warning saying when it will try again, waits until then without asking sooner, and says when it is back. `run` started during this waits the same way instead of failing; `doctor` warns and says when to try again.
 - **`claude` stops working** (signed out, out of plan, or uninstalled): the review it was running fails with Claude's own message, and the Sifter stops taking new reviews so the rest of your queue waits for you instead of failing. It logs one line saying what to fix, checks again by itself (after 15 seconds at first, backing off to every 5 minutes), and starts claiming again once `claude` works. A check costs nothing while you are signed out, and one short turn once you are back.
 - **Sift needs a newer Sifter**: it finishes the review in hand, installs the newest version with `npm install -g`, and restarts itself. If the install fails, it exits with code 2 and prints the command to run.
 - **The pairing is revoked** (removed in Sift, or replaced by a newer pairing): it exits with code 3 and shows the two ways to pair again. A review it was running goes back to the queue for another Sifter.