CI reliability overhaul: kill phantom failures, unblock releases, gate real signal (0283) - #414
Merged
Merged
Conversation
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Signed-off-by: xNet Test <test@xnet.dev>
mobile-e2e.yml had an unparseable plain scalar (': ' inside the run: echo
lines), so GitHub created a phantom failed run on every push to every
branch (48/48 failures). Quote both lines.
xnet-audiotee declared .macOS(.v14) while calling
AudioHardwareCreateProcessTap (14.2+), failing every Electron Release
macOS build since 0279 merged. Raise the deployment floor to 14.2;
runtime capability selection already gates on darwin >= 23.4 (macOS
14.4) in core-audio-tap.ts before probing the helper, so no runtime
change is needed. Verified locally with swift build -c release.
Signed-off-by: xNet Test <test@xnet.dev>
Fallow was the pipeline's slowest job (~17 min avg), failed 44% of PR runs on advisory notes despite never being a required check, and its concurrency group cancelled 6 of the last 27 runs mid-flight. Remove the pull_request trigger (user decision — stronger than the report-only option in exploration 0283): the weekly scheduled run and manual dispatch keep the hard --fail-on-issues gate and SARIF upload. Also re-enable the turbo cache and simplify the now-single-base ref plumbing. Signed-off-by: xNet Test <test@xnet.dev>
…a PR visual-capture.yml documents itself as informational but a Build Storybook failure still failed the capture job — a red X on 29% of recent runs. Add job-level continue-on-error to honour the contract. The real signal it carried (the storybook plugins shim drifting from what the story bundle imports out of @xnetjs/plugins — broke three times across 0279/0280/#412) moves into the required lint job: check-storybook-shim.mjs walks the story import graph with the same alias map storybook uses and fails in seconds when a name is missing from the shim. Backfills the one current gap (ImporterContribution, type-only so the build never noticed). Signed-off-by: xNet Test <test@xnet.dev>
One unparseable workflow file produced 48 phantom failed runs (0283); nothing validated .github/workflows. Add actionlint 1.7.12 to the required lint job and a guarded pre-commit pass for staged workflow files (advisory when the binary is absent). shellcheck severity floor is 'warning' in both so local and CI agree. Fix what it flags at that level: ai-changeset.yml inlined the attacker-controlled PR head ref into a script (moved to an env var), schema-check.yml had unquoted command substitutions inside JSON echoes, and three retry loops used an unused loop variable. Signed-off-by: xNet Test <test@xnet.dev>
The dco.yml check failed on 22 of 29 recent runs because agent-driven commits never carry Signed-off-by (0283). prepare-commit-msg adds the trailer from the configured git identity — dependency-free so it works in worktrees without node_modules, idempotent, and untouched by --no-verify (git only skips pre-commit/commit-msg). The CI check stays as the backstop. Signed-off-by: xNet Test <test@xnet.dev>
The daily rebuild pushed directly to main, which the repo ruleset
rejects ('changes must be made through a pull request') — the cron
failed every day the index changed (0283). Refresh a standing
plugins-registry/data PR via peter-evans/create-pull-request, labelled
skip-changelog and signed off; merging stays deliberate, same as the
changesets release PR (and with the same no-CI-on-GITHUB_TOKEN-PRs
admin-merge caveat).
Signed-off-by: xNet Test <test@xnet.dev>
Native Helpers workflow (path-filtered, macos-14) compiles the audiotee Swift helper on PRs that touch apps/electron/native/**, so macOS availability drift surfaces before release time instead of after three failed release runs (0283). assert-fragment.mjs mirrors the changesets Stop hook: when the branch changed app/package source without a changelog fragment, nudge once at turn-end (the required changelog-section check would otherwise reject the PR with feedback arriving only at PR time). Deliberately one-shot and skippable — skip-changelog is a legitimate answer. Signed-off-by: xNet Test <test@xnet.dev>
Signed-off-by: xNet Test <test@xnet.dev>
Signed-off-by: xNet Test <test@xnet.dev>
The dispatch-only stub never built @xnetjs/* deps, so build:web died on TS2307 the first time the (now-parseable) workflow actually ran. Signed-off-by: xNet Test <test@xnet.dev>
Contributor
🖼️ UI changes in this PRNo UI changes detected in this PR. |
Signed-off-by: xNet Test <test@xnet.dev>
Signed-off-by: xNet Test <test@xnet.dev>
Contributor
|
Preview removed for PR #414. |
crs48
added a commit
that referenced
this pull request
Jul 8, 2026
#416) Follow-up to #414 (exploration 0283). The 14.2 deployment-floor fix got audiotee compiling again on the Electron Release macOS jobs — which exposed the next line of the same step: SwiftPM symlinks `.build/release` to the arch directory, so the normalising `cp` copied the binary onto its own inode and exited 1 on both arches. Skip the copy when source and destination are the same file (`[ "$SRC" -ef "$DST" ]`). Validated via branch dispatch (run 28967007377): both `build-macos` jobs now pass `Build xnet-audiotee helper`. Also checks off the Electron Release and Plugins Registry validation boxes in the 0283 doc (registry validated by dispatch → PR #415, merged). 🤖 Generated with [Claude Code](https://claude.com/claude-code)
crs48
added a commit
that referenced
this pull request
Jul 8, 2026
## Summary The dispatch-only `mobile-e2e.yml` workflow (exploration 0238, un-broken in #414) failed at **Add Android project**: `apps/mobile` had no Capacitor dependencies and every script shelled out to `npx --yes @capacitor/cli@latest`, but Capacitor 8's `cap add android` requires `@capacitor/android` installed in the project. - Pin `@capacitor/cli`, `@capacitor/core`, `@capacitor/android`, `@capacitor/ios` at **8.4.1** as devDependencies of `xnet-mobile-shell` (private, no changeset needed). - Switch all scripts from npx-latest to the pinned CLI (`pnpm exec cap …`) — reproducible and no surprise majors. - Update the README prerequisites section, which documented the npx approach. ## Verification `gh workflow run mobile-e2e.yml --ref claude/upbeat-napier-8b894c -f platform=android` → [run 28968475239](https://github.com/crs48/xNet/actions/runs/28968475239) **success**: every step green through the `Run Maestro flows` TODO, including `[success] android platform added!` and Capacitor sync. Emulator/Maestro wiring remains the documented TODO. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Implements docs/explorations/0283 — CI failure patterns and pipeline health.
Over the last ~500 Actions runs, ~75% of failures came from four causes unrelated to code correctness, while the one real breakage (Electron Release) went unnoticed in the noise.
Tier 1 — stop the bleeding
run:lines whose:made the file unparseable — GitHub emitted a phantom failed run on every push (48/48 failures, ~40% of all failure volume).Package.swift: deployment floor.macOS(.v14)→.macOS("14.2")(AudioHardwareCreateProcessTapis 14.2+). Every Electron Release macOS build since 0279 failed on this; verified locally withswift build -c release. Runtime capability selection already gates on darwin ≥ 23.4 before probing the helper.--fail-on-issuesgate; turbo cache re-enabled. It was the slowest job (~17 min) and failed 44% of PR runs on advisory notes.continue-on-errorso the self-described informational check can't red a PR.Tier 2 — prevent the classes
lintjob + guarded pre-commit pass (shellcheck severity floorwarningin both). Fixed what it flags: untrustedhead.refinlined in ai-changeset.yml (moved to env), unquoted command substitutions in schema-check.yml, unused loop vars..husky/prepare-commit-msg: dependency-free DCO auto-sign-off (survives--no-verify; merges/squashes untouched). DCO failed 76% of recent runs.scripts/check-storybook-shim.mjs(pnpm check:storybook-shim): walks the story import graph with storybook's own alias map and fails thelintjob in seconds when the plugins shim misses a name — the drift that broke Visual UI Capture in 0279, 0280, and feat(web): labs toggle + workspace editing affordances (0282) #412. BackfillsImporterContribution.plugins-registry/dataPR (signed off,skip-changelog) instead of pushing to protectedmain, which repo rules rejected every time.Tier 3 — ratchets
macos-14swift buildon PRs touchingapps/electron/native/**, so availability drift surfaces pre-release.scripts/changelog/assert-fragment.mjsStop-hook nudge (mirrors the changesets hook): one-shot turn-end reminder when app/package source changed without a changelog fragment.Deferred (calendar-gated, unchecked in the doc)
Expected outcome: failed-run rate drops from ~21% to ~2-3%, and a red ✗ regains meaning.
🤖 Generated with Claude Code