Why is your CI so slow?Find its fastest descentwithBoat.dev
Independent hackathon project built on Boat.dev, not an official Boat.dev product.
Paste a public GitHub repo and Boat draws your CI and its critical path. Then one sentence in Claude Code sends your agent after the fastest path down.
The problem
Agents write faster. CI is the wait.
Anthropic reports its CI job volume grew 25× in six months (source). Teams on X describe the same wall, in their own words.
nowa@nowazhuWhat CI service is everyone using now? Since my agents started making more than 1,000 commits a day, I have spent USD 4,500 on GitHub Actions for CI. I set up a local runner myself, but its concurrency is limited too. This is now seriously limiting PR merges. · Translated from Chinese60 replies12 reposts69 likes38.6K viewsView on X
owen@owengretzingerHow I made our CI 20x faster (steal this)A year ago, a PR at Boardy waited up to 40 minutes for CI. Now it waits 2. That matters more than ever, because every PR we ship is written with coding agents, and agents don't get tired of pushing.130 replies240 reposts2.7K likes838.1K viewsView on X
Kun Chen@kunchenguidnow that we all agreed to not review most of the code any more, CI became the next bottleneck
at agent speed, if our CI pipeline is simply "run all the build, lint, tests", it'll become a major slow down …101 replies55 reposts797 likes68.5K viewsView on X
Daniel Lockyer@DanielLockyerIf you're working on making your CI faster and want your agent to visualize the timeline of jobs for you, give it a GitHub Actions link and ask it to output a .pftrace file that you can upload to Perfetto41 replies34 reposts733 likes50K viewsView on X
Counts are the visible X counters on Nov 20, 2026. Long posts are cut with an ellipsis; an article shows its title and opening lines. An independent project built on Boat.dev, not an official Boat.dev product.
What you get
A skill. A database. The experience. The infra. One prompt.
Everything a coding agent needs to make your CI faster, distilled from what a fast-moving public repo actually did, measured. Paste one sentence and it starts on your repo.
1 A skill
The playbook, written for agents
8 CI speedup practices: when each applies, what to change in the workflow, and the formula behind its estimate. skills.md
2 A database
Your repo's CI, as data
The pipeline, the measured timings and a plan with estimated savings, for your specific repo, from one API call.
3 The experience
What worked, and what did not
17 real CI changes from T3 Code, measured before and after: 7 followed by a faster run, 5 by a slower one.
4 The infra
Boat.dev machines, in parallel
Where the trial runs execute. Measured, one run each: four pandas test directories took 86.6 s on one Boat.dev machine and 17.2 s on eight. Planned, not live: Boat starting the trials itself.
5 One prompt
Paste it. It finds the path.
Into Claude Code or any coding agent, inside your repo. The agent is yours; Boat supplies the rest.
Read https://chaorders.com/boat/skills.md and make this repo's CI faster.Public GitHub repositories on GitHub Actions. Every number is labelled measured or estimated.
The experience, good and bad
Real changes, measured.
Your agent reads every change the T3 Code team made to its CI, with the median run before and after. Some made it faster. Some did not. Both are in the skill database. One source repository today; more are next.
Measured Followed by a faster run · 7
| Change | Before | After |
|---|---|---|
| Parallelize tests, split out Rust checks | 3m 36s | 1m 30s |
| Boot the macOS lint runner only for native changes | 3m 35s | 1m 30s |
| Split the single job into parallel jobs | 3m 57s | 2m 14s |
| Split web tests over two runners | 4m 4s | 2m 57s |
| Run server test files in parallel on each shard | 3m 53s | 2m 54s |
Measured Followed by a slower run · 5
| Change | Before | After |
|---|---|---|
| Move the transfer report job to Blacksmith | 2m 14s | 2m 54s |
| Cache server tests individually across shards | 3m 2s | 3m 38s |
| Cancel stale runs, drop redundant setup | 3m 18s | 3m 37s |
| Run PR checks side by side, balance server shards | 2m 9s | 2m 26s |
| Shards survive a throttled task cache | 3m 30s | 3m 43s |
Measured: median main-branch CI run of pingdotgg/t3code in the 7 days before and after each change to ci.yml. The repo and its tests kept growing, so a slower run after a change does not prove the change caused it. 5 more changes have no before measurement. Not live yet: more source repositories, Boat running the trials itself, and CD.
Already analyzed
Popular repositories, mapped.
Open one to see its pipeline, the one-sentence explanation, and the recommended plan.
pingdotgg
t3code
New and fast-moving: an open-source coding agent app.
- CI map
- ready to open
openclaw
openclaw
New and very popular: a personal AI assistant with a large CI workflow.
- CI map
- ready to open
apache
kafka
Since 2011 and still pushed daily. Its JUnit jobs run for hours without test shards.
- CI map
- ready to open
pandas-dev
pandas
Since 2010 and still pushed daily. Dozens of unit test jobs, the longest close to an hour.
- CI map
- ready to open
Read from each project's public workflow files and public GitHub Actions run history. Numbers marked estimated come from a formula shown on the page, not from a run on Boat. These projects are not affiliated with Boat and have not endorsed it.
Find your CI’s fastest descent.
Paste this into your coding agent
Read https://chaorders.com/boat/skills.md and make this repo's CI faster.Public repositories only. Boat reads the workflow files and the public run history, nothing else.