Diagnosed daily CI failure: only CompatHelper workflow failing
(since Sep 1), all other workflows green
Root cause: juliaregistries/compathelper action repo
was deleted upstream (404, not archived) → "Unable to
resolve action" at ~7s every run
Checked upstream: CompatHelper.jl is deprecated ("maintenance mode —
migrate to Dependabot", which now supports Julia natively)
Verified XAct.jl is a clean General-registry package (only
non-General manifest entry is the normal SxAct
self-reference) → safe for Dependabot
Fix (commit f5fd1a8, pushed):
Deleted .github/workflows/CompatHelper.yml
Added julia package ecosystem to
.github/dependabot.yml (weekly,
dependencies/julia labels, deps:
prefix) matching existing github-actions/pip entries
Committed with explicit file paths (per AGENTS.md rule),
git push succeeded, branch up to date with origin
Next: confirm Dependabot opens Julia update PRs on
next scheduled run; caveat — Dependabot doesn't support custom/private
registries yet, revisit if one is added
Session:15:38
(Testimonial.jl — CI failure diagnosis + fix)
Diagnosed failing CI across 3 workflows (gh run logs): all stem from
scripts/TestimonialRunner/ never being instantiated on CI —
Testimonial is unregistered, Manifests git-ignored → registry resolution
error (nightly) or driver subprocess exit 3 → record_item
returns nothing (E2E/ci.yml 1.11 jobs)
Ran Rule-of-5 review (rule-of-5-universal skill) on proposed fix;
caught 3 gaps: missing root Pkg.instantiate() in nightly,
replace-not-append the broken step, stale step name. Verdict READY after
amendments
Fix (commit 219abdb): nightly-record.yml — replace
broken step with
Pkg.develop(path=repo-root)+resolve+instantiate, add root
instantiate; ci-e2e.yml — ungate runner setup from Julia 1.12, rename.
actionlint clean, ah check clean
Verified via workflow_dispatch: nightly setup steps
pass (was dead at 24s); E2E 1.11 record_item tests now
pass; 1.12 still green
Closed testimonial-4ec8; filed 3 P1 follow-ups:
nmih (test_lcov_tracefile missing VERSION gate — masked by
old failure), l7hk (nightly record_all 30-min
timeout, 185 items, zero progress output — measure before re-budgeting),
ci.yml quick-suite 1.11 gate (red since Aug 5, same
defect)
Next:nmih (1-line test gate) → ci.yml
ungate → l7hk (measure per-item latency, then re-budget
timeout)
Fixed live-code language switching: hljs 11
data-highlighted marker silently skips re-highlighting →
clear marker in setLang
Added julia + clojure language modules (not in hljs common bundle) +
six new pills: css/html/xml/json/yaml/toml (html→xml alias, toml→ini
alias; no extra CDN deps)
Rule-of-5 review of index.html CTAs: hero had zero CTAs above the
fold (1280×800 headless evidence). Added hero CTA row
(Install/inspector/gallery), #install anchor, copy-to-clipboard on
install commands w/ execCommand fallback, corrected install copy against
actual install.sh behavior, Links descriptors
All verified headless via flatpak chromium (DOM dumps +
screenshots); 7 commits pushed, every Pages deploy green
Next: nothing queued. Ideas parked: /tree/ + gist
URL support in inspector, scroll-margin on #install anchor