More Than 200,000 Checks

Taskless just passed 200,000 checks since we started tracking them in May. That's every time someone ran taskless check against a codebase and got an answer back.
A good share of those checks are ours, and we're glad they are. We run Taskless on everything we build, and if it didn't work, we'd be the first to know.
We check ourselves
The Taskless product runs taskless check in CI, against the rules our own review history produced. The CLI's test suite runs checks against a test repository, which is how we find out a release broke something before you do. And every time we generate a report on a public repo, like the ones we did for Sentry and n8n, that's more checks.
Our writing goes through taskless check too. Consistent writing tone is hard across marketing copy, documentation, code comments and everything in between. Having a consistent voice and tone means everything Taskless does is shown the same way across every surface, whether it's marketing or code.
That's the same loop we built Taskless for. A reviewer keeps leaving the same comment, the comment becomes a rule, and the rule runs before anyone has to leave the comment again. It works on prose for the same reason it works on code: once the rule exists, nobody has to remember it.
What outside teams taught us
Since late August we've been able to tell which GitHub organization a run comes from. 27 outside teams have run Taskless in that time, for 47,068 checks. A few things stood out.
New rules start as warnings
When teams put Taskless in CI, they don't let it fail the build. Of the 10,067 checks that ran in CI, 3 reported an error. Most of the rest reported warnings, which show up in the log without blocking the merge. One team accounts for most of that CI volume, so take the ratio with that in mind. The other CI teams look the same, since those 3 errors are all there were across every team.
Locally it's the other way around. About 13% of local checks found an error, because that's where the agent is working and the rules are doing their job in the loop. The agent hits the rule, fixes the code, and runs it again.
We read that as trust being earned. A rule shows up as a warning in CI, the team watches it for a while, and once it stops being noisy it can become an error.
Teams start with a handful of rules
The median check runs about 4 rules. That matches our onboarding flow, which only tries to create a couple of rules from pull request comments and what it finds in AGENTS.md files. Even when a team shows up with a policy library, we help them start small. Nobody wants 40 rules showing up all at once.
The reports show the same thing. Pointed at Sentry's review history, Taskless wrote 10 rules, and those 10 found 253 places to fix (the Sentry report). On n8n, 5 rules found 1,897 (the n8n report). A short list of the right rules goes a long way.
CI happens in week one, or later
11 of the 27 teams have run Taskless in CI, and 5 of them did it in their first week. 16 teams came back in a second week. The teams that wire it into CI early tend to stay, which is the clearest signal we have that the rules were worth keeping.
What's next
We're going to keep counting, and we're going to keep counting ourselves in. If you want your repo's review history turned into rules, npx @taskless/cli is the place to start, and it takes about 10 minutes.