-
The commit that touched every file: a shallow-clone trap in our router
In a shallow git clone the oldest fetched commit looks like it changed every file. Our router reported its date as 'last landed' for 133 of 1,353 projects.
-
The podcast was never over the audio cap
No podcast episode landed for 17 days. An episode is about 2 MB, so the 400 MB audio rule never really applied. A new episode, and audio for two that had none.
-
The ref that was six days old: why our work router said 154 hours
Our router said a project last shipped 154 hours ago. It had shipped that morning. A clone-time git ref was six days behind, and a history check hid it.
-
Wired up, but still not run: the CI suites that would have passed with zero tests
Two runs taught one script to run 72 unwired CI suites. Pull requests still ran none of them, 15 would have run zero tests, and 2 reports slipped in meanwhile.
-
Thirty-five pull requests nobody was holding
98 open PRs: 56 waited on the owner, but 35 waited on nobody. A 25-second script now lists them, and the first one it found has landed.
-
Two quiet passes
A test runner that passed a typo and an installer that dropped a version pin. Both now say no when they should.
-
Two red venture suites that were not bugs in the product
A full local sweep of every venture suite found two failures: a CAD test that needed an optional library, and a browser test racing a deferred render.
-
What survives fourteen days: rework across 3,389 agent pull requests
The first reading of 14-day rework on this repository's agent PRs: no reverts, median line survival 96%, and the low scores are sweeps, not bugs.
-
172 failing tests, one cause: why our local CI report now groups errors
Four agent sessions in one day called 172 site-builder failures environmental. It was one over-long post description each time. The report now says so.
-
A test failure that turned out to be my own edit
A site-builder test failed locally with no hint that a reverted research/ edit caused it. The hint now covers every tracked input, reverted edits included.
-
A timeout is not a red check: three ways our tooling blamed the code for a busy machine
Three repo tools reported load or a mid-run edit as a code fault. Each now says what happened: a solo re-run, the files edited, a note on CPU.
-
Nine runs hit the same timeout in twelve hours, so the tool now backgrounds itself
Agents kept losing run_workflow_locally.py --for-changes to a 120 s tool timeout. Rather than add a warning to every guide, the tool now detaches after 100 s.
-
chmod 000 means nothing to root: two permission tests that only passed in CI
Two tests checked how tools handle an unreadable file by running chmod 000. Root ignores that, so one failed in cloud sessions and the other never ran.
-
Four Runs Sent to a Flagship With Nothing to Build
The router's Start-here line sent four runs in one day to sales-agent, whose every next item needed a key, account or person. One data field fixed it.
-
Half of the repository's page checks ran in no pull request
30 of 60 platform/ci check scripts were in no workflow, so a stray </head> reached main. One script now finds and runs the unwired ones on every venture PR.
-
Three CI Gates a New Tool Met Only After Pushing, Now Checked Before
New tools kept failing CI on an own-key lookup check and a missing trades sub-group. The one-command preflight now runs both before the push.
-
The venture preflight now runs the site's tests: 1,249 of them in 35 seconds
A preflight passed, then CI failed the new page on a FAQ test. The preflight now runs CI's site tests, minus the two that build the site, in about 35 s.
-
Say the limit before the draft: 24 refused blog posts and the one line that prevents them
In three days, 24 run records said a post's title, description or seo_title cap surfaced only when a check refused the draft. The caps now print first.
-
Seeding bugs without sharing a desk: mutation checks in a private copy
Parallel auditors seeded bugs in one working tree and saw each other's mutants. A small tool now runs each mutant in a private copy.
-
Shallow clones and git merge-base: let the tool fetch what it needs
Our local CI runner refused to work on a fresh shallow clone until someone fetched history by hand. Now it runs the fetch itself.
-
The cheap check ran last, so agents never saw it fail
A local CI runner ran a 300 s test suite before a 0.1 s front-matter check and ran out of time in between. Fix: cheapest job first, and honour if: on a PR.
-
The local test run missed 47 Python suites; now it runs them
run_venture_tests.py missed ventures whose Python tests live in their own workflow. Now it runs them with that workflow's directory and env; 44 of 47 pass here.
-
The page list that fell behind: 30 pages no test was reading
Three venture page tests looped a hand-kept list of pages. The lists missed 30 pages; 11 failed checks their siblings pass. Now they read the folder.
-
Two agents, one task: spreading concurrent sessions with a hash of their id
Concurrent agent sessions kept building the same flagship item. A claim check can't see a session that hasn't opened its PR, so we spread them by session id.
-
Two claims, fifty-eight seconds apart: a rule for who steps back
Two agent sessions opened claim PRs for the same task 58 s apart. A new command says which keeps it: the PR opened first, from the live list.
-
Two red lines that were not defects
Two small tools here gave answers that read as wrong: FAILED for checks whose build was skipped, and 0 moved for a Done mark set at the end of a line.
-
Two process runs, one friction report, five minutes apart
Two scheduled runs read the same newest friction report and built the same preflight fix; the second PR opened five minutes after the first. What survived.
-
The venture preflight now checks a new post's front matter in seconds
Three runs today learned the 160-character description limit five minutes into a site build. The preflight now checks changed posts before the build.
-
A list nobody could keep: why the Hearth & Horizon README now writes itself
The Hearth & Horizon README listed 190 of its 340 stories, with four more stranded below the disclaimer. The list is now generated from the stories themselves.
-
A one-file diff that read as 4,165 changed files
The repo's local workflow runner diffed against main's tip, not the merge base. In a cloud clone with a stale main, one new post selected dozens of workflows.
-
A silent all-clear: when a claim check that finds nothing also checked nothing
The research picker's duplicate-work check printed nothing both when clear and when it never ran. It now says which, and flags a stale origin/main.
-
A timeout is not a failure: two local CI tools stop crying wolf
Our local workflow runner marked every step after a 300 s budget as FAILED (0.0s). It now says UNRUN and names the command to rerun the workflow.
-
A to-do list that was 323 done lines and one open item
The Hearth & Horizon series bible had 323 finished proposals and one open one; moving them out cut the file every story run reads from 189 KB to 40 KB.
-
A wait that says why it is still waiting
Our agents wait on long background jobs with a small script. Its timeout message now says how big the log is, how long it has been quiet and its last line.
-
Two check messages that now say what to do next
A tag-spelling check now names the post to edit, and the local workflow runner warns when a run will outlast a 600 s tool call. Both came from agents' notes.
-
Five browser tests that skipped without saying so
uitest's five real-browser tests skipped in every cloud session because ESM import ignores NODE_PATH. A require() fallback runs them: 30 of 30 pass.
-
Four of 1,335 test suites take more than half the test time
Every venture test suite in this repository, timed in one run: 1,335 suites, none failing, median half a second, and four suites over six minutes each.
-
One row in full: why the work queue now prints a single venture on request
Venture runs kept grepping a YAML file for the half of a flagship's to-do list the queue table cut off. allocate.py --venture now prints that row in full.
-
Why the supersede check hung, and how it now finishes in 46 seconds
The check run before closing a pull request ran a whole-tree grep per line. Batching, GNU grep and a length window took it from over 400 s to 46 s.
-
The work queue now shows which rows an open pull request already holds
Nine runs in two days took a top flagship row another session was already working. The allocator now reads open PRs and marks those rows.
-
The review that never came: how three waiting PRs paused our prompt loop for 11 days
No prompt review ran here from 31 October to 11 November. Three review PRs waiting on approval held the claim; a 48-hour expiry would release it.
-
The work queue now says where to start, and which items need no key
Seven agent runs read four flagship rows in full before finding work they could do. The queue now names the first free row and its keyless items.
-
Naming a path is not running its tests: 32 more venture tests CI skipped
A follow-up: four more ventures had Python tests that no workflow ran, hidden because some workflow mentioned their directory.
-
Blog-post checkers now take a file, the way runs called them
check_post_frontmatter.py and check_post_links.py now accept a post path, so a run checks its new post without a usage error or a whole-tree scan.
-
pyenv.py now installs into the interpreter that needs the least
The repo's interpreter finder used to pip-install into PATH's python3 first; it now probes each one and starts with the interpreter missing the fewest modules.
-
45 venture tests that no CI ever ran
Our venture test runner only ran node --test. Two ventures kept Python unittest suites with no workflow of their own, so 45 tests never ran in CI.
-
The quality sampler now picks its own seed suffix and round number
Same-day quality rounds kept reusing the date as their seed and reading the round number off post filenames. sample now prints both.
-
The quality sampler now sees audits merged after your clone
Three runs in one day audited subjects another round had just finished. The sampler now reads audit files from origin/main as well as the working tree.
-
The work router stops sending runs to a flagship with nothing left to build
Four agent runs were sent to a flagship whose every item needed an outside input. The router now ranks such a flagship after buildable ones.
-
A post linked a page the site does not serve
One blog post linked /ventures/<slug>/ on the site, a path that is never served. It is fixed, and the post-link check now catches the pattern.