feat(ci): give the review bot a severity scale and a tally

REVIEW.md said what blocks and what does not, but never how to mark a finding,
so every review invented its own shape and none carried a severity. It now
names the three markers the hosted Code Review service uses - Important, Nit,
Pre-existing - and keys them to what the pull request did rather than to how
alarming a defect looks alone: a defect it introduces or worsens is Important,
one it merely brought into view is Pre-existing and cannot be a reason to hold
it. Pre-existing was missing entirely, and checking what this panel emits means
reading far outside the diff, so those findings had nowhere to go except a
wrong Important or silence.

The volume cap said how many and never which. It now collapses a nit repeated
across files into one finding, prefers a nit in code the pull request wrote
over one in code it only moved, caps pre-existing findings at three, and states
that Important findings are never capped - a section listing two caps otherwise
reads as licence to trim what matters. The review opens with a tally so the
author sees the shape before the detail.

Two contradictions went with it. The file told the reviewer to skip what CI
enforces and then to check that a new i18n key reaches all 13 locales, which
i18n-dead-keys.test.ts pins in both directions - the rule moves to "Do not
report" with the reason. "Anything CI already enforces: npm audit" overstated
what runs; CI audits production dependencies at high and above, so a
dev-dependency advisory is out of scope by design.

The reviewer could not read its own CI. Only postgres-durable-first runs
against PostgreSQL, and XRAY_E2E_BINARY and XUI_SCALE_TEST are set by no job,
so a dialect or migration change can carry a wall of green while the paths it
touches never executed. That belongs to the verification bar, next to the rule
that a behaviour claim needs a file:line citation, and "CI passed" now needs a
run actually read. Also names the two house choices no linter defends: neither
golangci-lint nor oxlint rejects a testify or Tailwind import.

Both kinds of claim rot on a rename, so a test pins them the way
repo-context.md's claims are already pinned - the CI jobs REVIEW.md names must
exist in ci.yml, the skip gates it calls unset must stay unset, and the locale
count must match the directory.

The review itself moves from high to max effort, and the prompt records why it
names REVIEW.md at all: the code-review skill reads CLAUDE.md on its own but
not REVIEW.md, so dropping that clause would silently stop the file applying.
Drops a CLAUDE.md reference to tools/seedperf/, which no longer exists - the
review reads that file as project context, so a stale path there misleads it.

.github/workflows/claude-bot.yml

@@ -496,11 +496,14 @@ jobs:
           persist-credentials: false
       # Read-only: this job holds a write-scoped token, so building or running
       # anything out of pr-head/ would turn the review into a pwn-request.
+      # checkout v7 refuses a fork PR ref outright unless that risk is accepted
+      # here, and nearly every pull request to this repository is from a fork.
       - uses: actions/checkout@v7
         with:
           ref: refs/pull/${{ github.event.pull_request.number || github.event.issue.number }}/head
           path: pr-head
           persist-credentials: false
+          allow-unsafe-pr-checkout: true
       - uses: anthropics/claude-code-action@v1
         with:
           github_token: ${{ secrets.GITHUB_TOKEN }}
@@ -508,13 +511,18 @@ jobs:
           allowed_non_write_users: "*"
           plugin_marketplaces: "https://github.com/anthropics/claude-code.git"
           plugins: "code-review@claude-code-plugins"
-          prompt: "/code-review:code-review high --comment ${{ github.repository }}/pull/${{ github.event.pull_request.number || github.event.issue.number }}"
+          # The skill reads CLAUDE.md on its own but NOT REVIEW.md - that file
+          # reaches a review only through the append-system-prompt below.
+          prompt: "/code-review:code-review max --comment ${{ github.repository }}/pull/${{ github.event.pull_request.number || github.event.issue.number }}"
+          # allowedTools only pre-approves; it denies nothing. Only the deny
+          # list stops the review executing what it just checked out.
           claude_args: |
             --model claude-opus-5
             --effort xhigh
             --max-turns 100
-            --allowedTools "mcp__github_inline_comment__create_inline_comment,Bash(gh api:*),Bash(gh pr diff:*),Read(//tmp/**),Write(//tmp/**),Edit(//tmp/**),WebFetch,WebSearch"
-            --append-system-prompt "Before reviewing, read REVIEW.md at the repository root and follow it: it defines what counts as a blocking finding in this repository, what not to report, and the repo-specific checks. Five overrides apply here. First, the skip gate for already-reviewed PRs: an existing Claude review comment justifies skipping ONLY when its 'Reviewed head:' SHA equals the PR's current head SHA; when the head has moved on, or this run was triggered by an explicit '@claude review' comment, run the full review, focusing on the commits since the previously reviewed head. Second, this is a headless run that terminates the moment you end your turn: launch every subagent with run_in_background set to false and wait for its result inside the same turn - never end your turn while a subagent is still running, and never end it before the review comment is posted. A run that ends without posting the review has failed. Third, the comment you post is the only part of this run anyone can see: it must carry the coverage list REVIEW.md asks for, whether or not you found anything. Fourth, the default working tree is the BASE branch, and a read-only checkout of the pull request head sits beside it in pr-head/: read and grep the changed files under pr-head/, and treat anything read outside it as the pre-merge baseline rather than as the code under review. Never build, install or execute anything from pr-head/ - this job holds a write-scoped token, so running pull-request code with it is the workflow vulnerability REVIEW.md itself calls blocking. Fifth, you cannot build or test here, but CI already did: read the head commit's checks with 'gh api repos/OWNER/REPO/commits/HEAD_SHA/check-runs' and report what they actually concluded instead of writing that verification was unavailable. A required check that failed, or that never ran on this head, is itself a finding."
+            --allowedTools "mcp__github_inline_comment__create_inline_comment,Bash(gh api:*),Bash(gh pr diff:*),Bash(grep:*),Bash(rg:*),Bash(ls:*),Bash(find:*),Bash(sed:*),Bash(git log:*),Bash(git show:*),Bash(git diff:*),Bash(go doc:*),Bash(go env:*),Read,Glob,Grep,WebFetch,WebSearch"
+            --disallowedTools "Bash(go build:*),Bash(go run:*),Bash(go test:*),Bash(go generate:*),Bash(go install:*),Bash(make:*),Bash(npm:*),Bash(npx:*),Bash(pnpm:*),Bash(yarn:*),Bash(node:*),Bash(bash:*),Bash(sh:*),Bash(docker:*),Bash(chmod:*),Edit,Write,NotebookEdit"
+            --append-system-prompt "Before reviewing, read REVIEW.md at the repository root and follow it: it defines the severity marker every finding carries, what counts as Important in this repository, what not to report, and the repo-specific checks. Five overrides apply here. First, the skip gate for already-reviewed PRs: an existing Claude review comment justifies skipping ONLY when its 'Reviewed head:' SHA equals the PR's current head SHA; when the head has moved on, or this run was triggered by an explicit '@claude review' comment, run the full review, focusing on the commits since the previously reviewed head. Second, this is a headless run that terminates the moment you end your turn: launch every subagent with run_in_background set to false and wait for its result inside the same turn - never end your turn while a subagent is still running, and never end it before the review comment is posted. A run that ends without posting the review has failed. Third, the comment you post is the only part of this run anyone can see: it must open with the tally and end with the coverage list REVIEW.md asks for, whether or not you found anything. Fourth, the default working tree is the BASE branch, and a read-only checkout of the pull request head sits beside it in pr-head/: read and grep the changed files under pr-head/, and treat anything read outside it as the pre-merge baseline rather than as the code under review. Never build, install or execute anything from pr-head/ - this job holds a write-scoped token, so running pull-request code with it is the workflow vulnerability REVIEW.md itself calls blocking. Fifth, you cannot build or test here, but CI already did: read the head commit's checks with 'gh api repos/OWNER/REPO/commits/HEAD_SHA/check-runs' and report what they actually concluded instead of writing that verification was unavailable. A required check that failed, or that never ran on this head, is itself a finding."
       - name: Upload the run transcript
         if: always()
         env: