[harlan-zw]

🔗 Linked issue

❓ Type of change

• 🧹 Chore

📚 Description

Every push of a PR re-runs lint, typecheck, and test in a fresh container, even when a push touches one markdown file. The pool is getting a shared cache volume (hogwild-gh-runner#10) and pnpm pipeline can use it: an unchanged task restores its outputs and replays its log instead of re-running.

The gate now runs with task-cache: true. harlan-nuxt#159 merged as 75ff091, so the pin is back on @main; the feature-branch pin existed only to prove this gate change end to end before it landed. The graph lives in pnpm-workspace.yaml: dev:prepare declares .nuxt/** outputs with config-only inputs, so the 30-90s Nuxt prepare restores on every push that does not move the app tree, and vitest keeps its .nuxt/tsconfig.json without the old prepare-command step.

pnpm moves to 12.4.2, the first line that accepts task settings. The lockfile gains the pnpm self-dependency block and nothing else moves. Lint task carries no inputs on purpose: eslint . reads the whole repository, so narrowing its key would replay a pass over files lint never saw.

Ran the test pipeline locally before pushing: 272 files, 1947 tests passed cold, then the second run restored dev:prepare and test from cache and replayed. The cache hits only start counting once the new supervisor is installed on Hogwild and the desktop; until then this runs as a normal cold gate.

🤖 AI disclosure: Harlan Agent Kit modified this description.

[harlan-github-agent]

🤖 MERGED

Harlan Agent Kit posted this automated review. It is not Harlan's personal review or approval. AI open source policy. Last updated: 2026-09-22 03:04 UTC.

GitHub merged this pull request.

• A PR that touches only workers/artifact-signer gets a green test job that runs no tests at all. Next: Preserve the base guarantee for the test job: pass --full to pnpm pipeline test on pull_request events too (task caching still replays unchanged tasks, so the savings largely remain), or re-add the worker suite as a pipeline task whose vi

182330b4-13c1-4d02-a9d7-e71ee65b7675

[harlan-zw]

Checked by hand, since CI cannot show it (the pool has no cache volume until the new supervisor is installed):

• pnpm pipeline test cold, local: 272 files, 1947 tests passed through the graph
• Same command again: dev:prepare and test restored from cache, logs replayed, no re-run
• XDG_CACHE_HOME relocation verified, which is the exact mechanism the runner volume uses

First CI run caught the artifact-signer suite failing for want of root .nuxt; that is the cross-project ordering limit described in pnpm-workspace.yaml, not a flake.

[harlan-zw]

Checked by hand for the review findings, since CI cannot show either:

• Finding 1 (stale green test replay) is real: 24 suites import scripts/ or checks/, so a scripts-only push left the test key unchanged. Both trees now hash into the test and typecheck keys.
• Finding 2 (stale typecheck replay) does not reproduce as stated: with a deliberate type error planted in scripts/, nuxt typecheck exits 0, so vue-tsc does not reach that tree today. The glob is in typecheck inputs anyway, because the generated tsconfig that sets its reach can drift, and a false miss costs seconds where a false hit ships a bug.

🤖 AI disclosure: Harlan Agent Kit wrote this comment.