The fleet's 22:09 UK incident (a Mac-side pkill -f <log file name> matched nothing, the roll-everything script lived on and wiped a held box) and the day's two pgrep self-matches are one class. The check flags pgrep -f / pkill -f with a plain literal (every one on a line), any pgrep/pkill on a file-name shape, and ps | grep with a literal; it allows the bracket form, -x, -F pidfile, kill $(cat pidfile), a variable and a full path; 11 banned and 16 allowed shapes in its self-test; 0.15 s over the tree. The 25 pkill -f sp1-gpu-server inside bash -c bodies (which matched the calling bash) are pkill -x; the other 11 literals take the bracket form; prover-socket-check accepts both. Row R in the record; the CLAUDE.md rule names the check and covers pkill and file names.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The box's 34 red rows of 6 October classified (docs/analysis/ci-failures-2026-10-06.md section 6): 22 iterations, 12 in three real classes (instant deaths with nothing kept, a shared worktree directory, an unread dry run). remote-run.sh now: pre-flight (subcommand, manifest, -p package, --features) refuses in a second with exit 3 and a class; the last 400 lines of every run kept in /srv/builds/_log/runs; a class on every row (compile-error, link-error, test-failure, instant, no-test-matched, slot-timeout, no-dir, preflight-*); a cargo test whose filter matched no test exits 3; a per-worktree lock in checkout and run mode; every red row appended to /srv/ci-red/red.jsonl as source box. red-watch.mjs never posts a box row alone and sends one digest a day (counts per class with each class's guard); the timer runs tick. Shared group cired on the box so the runner and build append to one file. Shown in a sandbox on the box: pass, failing test, empty filter, bad package, bad feature, missing subcommand, compile error, broken manifest, two concurrent runs of one worktree.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
pgrep -x git over the whole machine kept the lock whenever any git ran (the pre-push hook's own git push, another agent's build) and the checkout died with "index.lock: File exists". The fact is the lock's age. Class Q in docs/analysis/ci-failures-2026-10-06.md with the GIT_DIR leak of the previous commit.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
168 non-green runs since the first workflow run were classified (docs/analysis/ci-failures-2026-10-06.md): 102 were tree checks that finish in under 25 s on the pushing machine, 40 were GitHub-side refusals nobody saw.
tools/ci/pre-push.sh is the one list of fast checks; ci.yml's site job calls it with --ci and the pre-push hook with --hook (full gate for master and release-*, structural checks for other refs; never writes into the worktree). tools/ci/export-exclude.txt lists research documents outside the public export list, pruned by identity-check.sh and by the mirror's sync.sh (self-test: an excluded path may quote the patterns, an exported one may not); polish.md and this record are its first entries, which makes master green. tools/ci/windows-paths-check.sh (colon, trailing dot or space, reserved names, over 240 characters) runs as the pre-commit hook on staged paths and in the gate. tools/ci/red-watch.mjs plus the red job on the box's runner record one line per failed master or release-* run to /srv/ci-red/red.jsonl; igneum-ci-red.timer posts each once to the updates channel. pow and sims read IGNEUM_CI_RUNNER for the box. no-foreign-tree-writes.sh no longer exits silently on its warning pipeline under pipefail.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
