- review model: opencode-go/claude-haiku-5-5#medium instead of deepseek-v4.1-flash#max; flash has no medium variant - merge lint and typecheck into one job with a single pnpm install - cache the pnpm store and opencode binary via actions/cache@v4.1.0, keyed on a sha256 of the lockfile (hashFiles is empty on this runner) - update the decision records and log
4.1 KiB
type, title, description, tags, status, generated, verified, sources
| type | title | description | tags | status | generated | verified | sources | |||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Decision | PR pipeline layering | Three layers - pre-commit hooks for cheap deterministic checks, Gitea Actions for thorough checks, an AI reviewer for judgment - and the rule that decides what runs where. |
|
draft |
|
|
|
Decision
Every pull request is checked by three layers, ordered by cost:
- Pre-commit hooks (local, seconds, deterministic): gitleaks, file hygiene, conventional-commit check, actionlint, eslint --fix on staged files. A commit that fails here never exists.
- Gitea Actions (per PR push, minutes): ESLint + vue-tsc typecheck in one job (one install, cached pnpm store), gitleaks over full history, pnpm audit (blocking), droast Dockerfile lint, Snyk (currently disabled).
- AI code review (per PR push): opencode with a plan-covered model, verdict-based, posted as a PR comment.
The split rule: anything that finishes in seconds and auto-fixes goes in
pre-commit; anything slow, network-bound or quota-bound goes in CI; anything
that produces judgment rather than a pass/fail is the AI review. Nothing
slow lives in pre-commit — one slow hook gets bypassed with --no-verify
once and is dead weight forever.
Why
Each layer catches what the previous one cannot. Pre-commit sees the commit before it exists; CI sees the whole PR on a clean machine; the AI review sees intent, contracts and cross-file consequences that no linter can. The ordering also protects the budget: the AI review only runs for code that already passed the deterministic gates.
Caching
Two caches, both served by the runner's own cache server (no external service, entries stay on the home server):
- pnpm store, keyed on the store layout version (v11) and a
sha256sumofpnpm-lock.yaml. Installs hardlink from the store instead of downloading tarballs again. The key is built in a shell step becausehashFiles()evaluates empty in this act runner. - opencode binary, keyed on the pinned CLI version (
~/.opencode/bin). Small next to the model call, but it removes a network fetch per review.
Lint and typecheck also share one job: same image, same lockfile, same
install. Splitting them duplicated the install and made both compete for
the runner's two vCPUs. A future pnpm build job reuses the same store
cache key.
What the pipeline bought us in practice
Recorded so the reasoning survives: during the first week the pipeline
caught leaked secrets in history (gitleaks), a review model configured
against the wrong provider account, a Dockerfile-lint job that silently
never ran (hashFiles evaluating empty in the act runner), a review that
completed but exited non-zero, and an attacker-relevant credential
exposure in the review setup — each fixed where it was cheapest.
Failure modes deliberately designed against
| Failure seen | Guard now in place |
|---|---|
| Error logs posted as review comments | Posting gate is the output contract (a VERDICT: line), not the exit code |
| Vendor container images without arm64 manifests | Binaries installed arch-aware in node:22-bookworm, never vendor CI images |
hashFiles() silently empty in the act runner |
File discovery via find and cache keys via sha256sum, both in shell steps - never workflow expressions |
| Silent advisory failures | continue-on-error removed everywhere; a failed check is red |
| Supply-chain (curl-piped binaries) | Pinned versions, checksums pinned in the workflow |
Not verified
- That the Snyk job still works when re-enabled — it was disabled while the free quota was the binding constraint.
- That the pipeline behaves on Gitea runner versions other than v0.6.1.
Revisit when
- A second repository needs the same pipeline (extract a reusable workflow include).
- The Go plan's usage limits are hit by review traffic.