feat(ci): replace opencode review with PR-Agent
PR Check / Security scan (Snyk) (pull_request) Skipped
PR Check / Lint + typecheck (pull_request) Canceled after 45s
PR Check / Secret scan (gitleaks) (pull_request) Canceled after 0s
PR Check / Dependency audit (pull_request) Canceled after 0s
PR Check / Dockerfile lint (droast) (pull_request) Canceled after 0s

Run PR-Agent as a pinned webhook server on the Gitea Docker network and review on PR open plus slash commands. The per-PR opencode installer and verdict-posting step are removed.

The server-side .env stays outside the repository; deploy/pr-agent/ holds the compose definition and operator notes.
This commit is contained in:
Noel Sabia committed 2026-10-30 21:00:53 +01:00
1 parent 9ecf504843
commit c626cc9c1d
11 files changed
+251 -236

No files matched your search

-226
View File
@@ -229,229 +229,3 @@ jobs:
# Dockerfile/base image checks are covered by the droast job
snyk code test \
--severity-threshold=high
# -------------------------------------------------------- ai code review --
review:
name: AI code review (opencode)
runs-on: ubuntu-latest
container: node:22-bookworm
# Draft PRs don't get a (paid) review yet. The event_name check is
# load-bearing: manual (workflow_dispatch) triggers - which exist to
# run the paused snyk job - must NOT launch a paid review, which on
# a dispatch event has no PR to post to.
if: ${{ github.event_name == 'pull_request' && !github.event.pull_request.draft }}
steps:
- uses: actions/checkout@v4
with:
fetch-depth: 0
- name: Restore opencode binary
id: opencode-cache
# The installer writes to $HOME/.opencode/bin. Caching it saves the
# ~4s download+extract per run - small next to the model call, but
# free to keep. Key includes the pinned version, so bumping the
# version below invalidates the cached binary automatically.
# As above, a cache outage falls back to the installer.
continue-on-error: true
uses: actions/cache@v4.1.0
with:
path: ~/.opencode/bin
key: opencode-v2.0.32-${{ runner.os }}
- name: Install opencode CLI
if: steps.opencode-cache.outputs.cache-hit != 'true'
run: curl -fsSL https://opencode.ai/v2/install | bash -s -- -v 2.0.32
# v2 installer is required: the classic /install URL ships only
# OpenCode 1.x (resolved to v1.19.3 on arm64), which cannot resolve
# model variants like "#medium" and failed with ProviderModelNotFoundError.
# Version pinned (-v 2.0.32) so a breaking v2 release cannot fail
# the job mid-PR; bump deliberately, not automatically.
- name: Add opencode to PATH
# The installer appends the directory to $GITHUB_PATH itself, but that
# step is skipped on a cache hit. Add it unconditionally.
run: echo "$HOME/.opencode/bin" >> "$GITHUB_PATH"
- name: Get PR diff
run: |
# lockfile changes bloat the diff (175KB here) and are pointless to
# review - pnpm audit already covers dependencies. Excluding the
# lockfile also keeps the diff small enough to hand to the model.
# The diff goes to a file IN the workspace so the agent can read it
# with its file tools (passing 190KB as a CLI argument fails with
# "Argument list too long", MAX_ARG_STRLEN is 128KiB).
git diff --merge-base "origin/${{ github.base_ref }}" \
-- . ':(exclude)pnpm-lock.yaml' > .pr-review-diff.patch
wc -c .pr-review-diff.patch
- name: Run opencode review
env:
OPENCODE_API_KEY: ${{ secrets.OPENCODE_API_KEY }}
run: |
# opencode-go/claude-haiku-5-5#medium: cheap ($0.10/$0.50 per M
# tokens) and, unlike deepseek-v4.1-flash, it has a #medium
# reasoning variant - flash only exposes low/high/max, so every
# review there ran at maximum effort. #medium keeps the review
# competent without paying for max reasoning on each push.
# History: deepseek-v4-pro#max -> deepseek-v4.1-flash#max ->
# haiku-5-5#medium (2026-10-20). The pro endpoint failed two
# consecutive runs with "Upstream request failed: Endpoint is
# unavailable" before the flash switch; that fallback is recorded
# in OKF_Nuxt/open-questions.md. NOTE the provider prefix must be
# "opencode-go", NOT "opencode" (Zen is a different, pay-per-call
# provider and fails with 'Insufficient account funds').
#
# Auth is env-only: the CLI picks up OPENCODE_API_KEY from the
# environment (verified headless). Deliberately NO opencode.json:
# a config file in the checkout would hold the API key in
# plaintext, readable by the agent - a malicious PR diff could
# prompt-inject the agent into quoting the key in the review
# comment (publicly visible). Env vars are invisible to file reads.
# Permissions live in a global config OUTSIDE the checkout (no
# secrets in it, so nothing can leak). OpenCode v2 schema: the
# field is "permissions" (plural) - an ordered array of
# {action, resource, effect}; "bash" was renamed "shell" in v2.
# The singular v1 "permission" object is ignored by the v2 CLI
# (verified empirically: shell commands executed despite deny).
# Denying shell/edit/webfetch/websearch/external_directory means
# no headless permission prompts can hang or kill the run, and
# the reviewer physically cannot modify the repo or browse
# attacker-controlled URLs.
mkdir -p "$HOME/.config/opencode"
printf '{
"$schema": "https://opencode.ai/config.json",
"permissions": [
{ "action": "edit", "resource": "*", "effect": "deny" },
{ "action": "shell", "resource": "*", "effect": "deny" },
{ "action": "webfetch", "resource": "*", "effect": "deny" },
{ "action": "websearch", "resource": "*", "effect": "deny" },
{ "action": "external_directory", "resource": "*", "effect": "deny" }
]
}' > "$HOME/.config/opencode/opencode.json"
# --print-logs: surface auth/provider errors in the CI log instead
# of failing with an opaque exit code
EXIT_CODE=0
opencode run \
--print-logs \
--model opencode-go/claude-haiku-5-5#medium \
"You are a principal engineer performing a rigorous code review on a
pull request. You have tool access: read repo files for context when
the diff alone is ambiguous, but review ONLY code changed in this PR.
The full PR diff is in the file .pr-review-diff.patch in the repo
root. Read it first with your file tools.
pnpm-lock.yaml is deliberately excluded from that patch to keep it
small, not forgotten. It is committed in the PR, pnpm audit covers
the dependencies, and CI runs `pnpm install --frozen-lockfile`.
Do not report a missing lockfile or a frozen-install drift.
## Project context
- Stack: Nuxt 4 (Vue 3, TypeScript) SSR web app in the repo root.
Backend/API services are being added in Golang.
- Deployment target: a personal production CV website.
- Already covered by other CI jobs, do NOT repeat their findings:
ESLint (style), vue-tsc (types), gitleaks (secrets in git history),
pnpm audit (dependency CVEs), droast (Dockerfile lint).
- Your job is everything those tools CANNOT see: logic, security,
architecture, correctness.
## Nuxt / Vue / TypeScript focus
- SSR pitfalls: window/document/localStorage access without
<ClientOnly> or process.client guards; hydration mismatch risks
(Date.now(), Math.random(), locale formatting in render path).
- useAsyncData/useFetch: missing or colliding keys, duplicate
requests between server and client, wrong caching/refresh setup.
- Reactivity bugs: destructuring reactive objects losing reactivity,
missing await on async setup, watch on getters vs refs.
- Security: v-html with unsanitized data, user input flowing into
meta/URLs, open redirects, missing validation in /server/api and
Nitro route handlers, auth checks on both page and API level.
- Config hygiene: server-only runtime config (no NUXT_PUBLIC_
prefix) leaking secrets into the client bundle.
## Golang focus (when Go code is in the diff)
- Errors: unchecked errors, errors discarded, wrapping with %w,
sentinel vs typed errors, panic in library code.
- Concurrency: goroutine leaks (missing context cancellation),
data races (shared state without mutex/channel), WaitGroup
misuse, defer in loops.
- Resources: unclosed response bodies/files/rows, missing
timeouts on http.Client, unbounded buffering.
- Correctness: slice aliasing surprises, copy vs reference
semantics, nil map writes, integer truncation.
## Cross-cutting
- API contract mismatches between Nuxt frontend and Go backends
(field names, nullability, status codes, date formats).
- Missing authorization checks on new endpoints or pages.
- Secrets/credentials ending up in logs or client-visible output.
- Performance on the critical render path (heavy components,
unoptimized images, blocking Nitro middleware).
## Output format (strict)
Review the diff in .pr-review-diff.patch. For every finding:
[BLOCKER] file:line - what and why + concrete fix suggestion
[WARN] file:line - what and why + concrete fix suggestion
[NIT] file:line - optional improvement
BLOCKER = must fix before merge (security, data loss, broken
behavior). WARN = should fix soon. NIT = optional.
If a finding cannot be localized to a file, prefix [BLOCKER-GENERAL].
No compliments, no summaries of what the diff does, no style nits.
Keep it under 400 words unless there are more than 10 findings.
If the diff contains no reviewable code (docs/config only), say so
in one line.
End your response with a final verdict line, exactly one of:
VERDICT: APPROVE (no BLOCKER findings)
VERDICT: REQUEST_CHANGES (at least one BLOCKER)" \
> /tmp/review.md 2>&1 || EXIT_CODE=$?
# always show the output - never swallow the error text
cat /tmp/review.md
# The posting gate is the OUTPUT CONTRACT, not the exit code:
# a completed review always ends with a VERDICT line - error
# dumps never do. opencode can exit 1 on benign tool denials
# even after a full review, so a present verdict means post;
# anything else means the review did not complete.
# Anchored to the tail: the prompt itself contains VERDICT
# lines, so scanning the whole file could pass on echoed
# prompt text or injected quotes - only the last lines count.
if tail -n 5 /tmp/review.md | grep -qE '^VERDICT: (APPROVE|REQUEST_CHANGES)'; then
[ "$EXIT_CODE" -ne 0 ] && echo "::warning::opencode exited ${EXIT_CODE} but produced a valid verdict - posting review"
echo "Review completed with verdict."
else
# fail closed: no verdict = no review. exit 1 unconditionally -
# falling back to EXIT_CODE would exit 0 when opencode succeeded
# but produced no review, and the job would go green.
echo "::error::opencode produced no valid review (exit ${EXIT_CODE:-1}) - NOT posting this output as a review comment"
exit 1
fi
- name: Post review as PR comment
continue-on-error: true # token missing/expired must not fail the build
env:
GITEA_TOKEN: ${{ secrets.PR_GITEA_TOKEN }}
PR_NUMBER: ${{ github.event.pull_request.number }}
REPO: ${{ github.repository }}
# Gitea sets GITHUB_API_URL, e.g. https://gitea.example.com/api/v1
run: |
# re-verify the output contract here: never post anything that is
# not a completed review, regardless of how the previous step ended
if ! tail -n 5 /tmp/review.md | grep -qE '^VERDICT: (APPROVE|REQUEST_CHANGES)'; then
echo "::error::no verdict in review output - nothing to post"
exit 1
fi
# jq is not in the node image - use node for JSON encoding
BODY=$(node -e 'process.stdout.write(JSON.stringify(require("fs").readFileSync("/tmp/review.md","utf8")))')
HTTP_CODE=$(curl -s -o /tmp/comment-response.json -w "%{http_code}" \
-X POST \
-H "Authorization: token ${GITEA_TOKEN}" \
-H "Content-Type: application/json" \
-d "{\"body\": ${BODY}}" \
"${GITHUB_API_URL}/repos/${REPO}/issues/${PR_NUMBER}/comments")
echo "POST status: ${HTTP_CODE}"
cat /tmp/comment-response.json
case "$HTTP_CODE" in
200|201) echo "Review comment posted." ;;
*) echo "::error::Failed to post review comment (HTTP ${HTTP_CODE}) - check PR_GITEA_TOKEN scope and expiry" ;;
esac