/approve-architecture — Record Per-PR Design-Review Approval

Writes .claude/session/reviews/<owner>__<repo>__<pr>-architecture.approved (repo-qualified path, see AgDR-0060) with the current HEAD SHA so the require-architecture-review.sh merge-gate hook will let a design-artifact PR through. Without this marker, the hook blocks merges on any PR that touches a technical design, a migration AgDR, or a feature spec / PRD.

This skill is the architecture-review analog of /approve-design (UI gate) and /approve-merge (CEO gate). Same pattern, different gate. The reviewer is Tariq (the Solution Architect).

The one rule you must not break

INVOKE THIS SKILL ONLY ON EXPLICIT, PER-PR, DESIGN-REVIEW APPROVAL.

Normally Tariq writes the marker himself on an APPROVED verdict (see .claude/agents/solution-architect.md). This skill is the operator path to record the same marker — for when a human architect reviewed the design, or when you need to re-record after a rebase.

Valid invocation triggers:

• "design review passed" / "architecture approved" / "the design in #42 is sound" — if and only if the surrounding context clearly names a specific PR and the design has actually been reviewed against the architecture lens.
• "PR #42 architecture approved" — names the PR explicitly.
• A reply to your own "PR #42's design — architecture review approved?" message that consists of any affirmative token.

Invalid triggers (do NOT run this skill):

• "looks good" / "nice design" — when said about a whiteboard sketch, a Figma, or a verbal proposal that is not a specific PR's design artifact. Architecture review means reviewing the committed design doc / AgDR / spec in a PR, not a sketch.
• "the approach is fine" — when said in a planning context ("let's go with this approach") rather than a review context ("I've reviewed the design doc against the lens and it's sound").
• "go" / "continue" / "ship it" — umbrella responses to a multi-step plan. Same rule as /approve-merge.
• Your own inference that "the design is probably fine." NO. Stop and ask.

If in doubt: STOP AND ASK. "PR #X carries a technical design — has it been reviewed and approved against the architecture lens?" is one message. Building against an unsound design is much worse.

Process

1. Parse the PR number — and the repo

Extract the PR number from the argument. If none given, infer from the current branch's open PR (gh pr view --json number --jq '.number') or the user's most recent message. If ambiguous, STOP and ask.

Also resolve the repo (REPO). Accept the fully-qualified owner/repo#N form, or an explicit owner/repo second token. In split-portfolio v2 the PR lives in a sibling repo, so a bare gh pr view <pr> resolved against the ops-fork cwd hits the WRONG repo — the marker would then be written under the ops-fork qualifier and the require-architecture-review.sh gate (which keys on the PR's real repo, derived from the merge command's cd-target, me2resh/apexyard#687) would never find it → false-block. Pass --repo "$REPO" to every gh pr view call below when REPO is known. Fail loud: if only a bare number was given and gh pr view <pr> cannot resolve the PR from the current cwd, STOP and ask for the owner/repo#N form — never write the marker under a guessed qualifier.

2. Sanity-check the user's intent

Re-read the user's most recent message:

• Did they explicitly name this PR, or can I point at a direct "PR #X architecture approved?" question I just asked?
• Was the design artifact in the PR reviewed, or just a sketch / verbal proposal?
• Is the approval for this specific PR's design, or for a general direction?

If any are unclear — STOP and ask a per-PR explicit question.

3. Verify the PR state

gh pr view <pr> ${REPO:+--repo "$REPO"} --json state,isDraft,mergeable,headRefOid

• state must be OPEN. Refuse if MERGED, CLOSED, or DRAFT.
• mergeable should be MERGEABLE or UNKNOWN.
• Capture headRefOid — the marker must match the PR's GitHub HEAD.

4. Verify the Rex marker exists at current HEAD

Architecture sign-off is a stamp on top of a Rex-approved HEAD (the design PR still gets a normal code review for its prose / diff), not parallel to it. Resolve the ops fork root (NOT git toplevel — inside workspace/<project>/ the markers live in the ops fork above) and source the marker path helper:

REPO_ROOT=$(git rev-parse --show-toplevel)
OPS_ROOT=""
r="$REPO_ROOT"
while [ -n "$r" ] && [ "$r" != "/" ]; do
  if [ -f "$r/.apexyard-fork" ]; then OPS_ROOT="$r"; break; fi
  if [ -f "$r/onboarding.yaml" ] && [ -f "$r/apexyard.projects.yaml" ]; then OPS_ROOT="$r"; break; fi
  r=$(dirname "$r")
done
MARKER_HOME="${OPS_ROOT:-$REPO_ROOT}"
# shellcheck source=/dev/null
. "$MARKER_HOME/.claude/hooks/_lib-review-markers.sh"
# Base (host) repo — the canonical marker key: it matches solution-architect.md's
# architecture marker AND the require-architecture-review.sh gate's lookup (which
# keys on the merge command's base repo, #765). Prefer the repo resolved in step 1
# (already the base, #687); if it wasn't given, fall back to the CURRENT
# checkout's own remote — a deterministic, non-ambient source of truth. Do NOT
# fall back to an unscoped `gh pr view <pr> --json headRepository`: that call
# reads the wrong field (the PR's head/fork) and is itself an ambient-resolved
# gh query that can silently prefer the wrong repo in a fork checkout (#887).
# pr_base_repo now REQUIRES this repo and scopes its own gh query to it — never
# gh's ambient default — so same-repo PRs still resolve unchanged.
if [ -n "$REPO" ]; then
  REPO_FOR_BASE="$REPO"
else
  origin_url=$(git remote get-url origin 2>/dev/null)
  origin_url="${origin_url%.git}"
  REPO_FOR_BASE=$(printf '%s' "$origin_url" | sed -E 's#^(https?://[^/]+/|git@[^:]+:)##')
fi
PR_HOST_REPO=$(pr_base_repo <pr> "$REPO_FOR_BASE")
PR_REPO="$PR_HOST_REPO"
REX=$(review_marker_path "$PR_HOST_REPO" <pr> rex "$MARKER_HOME")
[ -f "$REX" ] && [ "$(tr -d '[:space:]' < "$REX")" = "<headRefOid from step 3>" ]

If Rex's marker is missing or its SHA doesn't match HEAD, refuse and tell the user to run the code-reviewer first. Do not write the architecture marker on a stale base.

5. Verify the PR actually carries a design artifact

Check whether the PR's diff includes files that trigger the architecture-review gate (technical design, migration AgDR, PRD / spec). If it has none, the marker is unnecessary — tell the user and skip.

gh pr diff <pr> --name-only | grep -qiE '(docs/agdr/.*migration.*\.md|technical-design|tech-design|/designs/|/prds/|prd.*\.md|feature-spec)'

6. Write the architecture marker

Use the repo-qualified path via _lib-review-markers.sh (already sourced in step 4):

# (MARKER_HOME and PR_HOST_REPO already resolved in step 4 — reuse them here.)
mkdir -p "$MARKER_HOME/.claude/session/reviews"
# architecture marker keyed on the BASE repo — same key as solution-architect.md
# + the gate (#765). Keying on the fork would leave a cross-fork design PR blocked.
ARCH=$(review_marker_path "$PR_HOST_REPO" <pr> architecture "$MARKER_HOME")
printf '%s\n' "<headRefOid>" > "$ARCH"

The file contains exactly one line: the 40-character HEAD SHA + newline. No labels, no JSON.

7. Confirm to the user

Architecture approval recorded for PR #<pr> at <sha>. The architecture-review merge gate will now allow this design PR through.

Do NOT run gh pr merge yourself. The skill's job ends at recording the marker. The merge is a separate action that still requires the CEO marker via /approve-merge plus an explicit merge instruction.

Notes

• The marker is gitignored (.claude/session/ is in .gitignore). Session state, not code.
• Re-running /approve-architecture <pr> is idempotent — overwrites with current HEAD.
• New commits after approval invalidate the marker (the gate compares SHAs) — re-request review.
• This skill does NOT invoke the Solution Architect role. It records approval after the design has been reviewed (by Tariq via /design-review, or by a human architect).

Anti-pattern

Architect: "The approach we discussed sounds right, go for it"
You: *tries to invoke /approve-architecture 42*  ← WRONG, twice over: a verbal
                                                  nod on an approach is not a
                                                  review of the committed design
                                                  artifact, AND since #1042 the
                                                  model cannot invoke this skill
                                                  at all.

A verbal approval of an approach is not a review of the committed design artifact. The correct flow:

Tech Lead: *commits the technical design to PR #42*
You: *runs /design-review so Tariq reviews it against the architecture lens*
... Tariq reviews, verdict APPROVED, and writes the marker himself ...
                                  ← DONE. No /approve-architecture needed.

Tariq writing the marker on an APPROVED verdict is the normal path, and it already satisfies the gate. This skill is the operator path for the other case: a human architect reviewed the design, or the marker needs re-recording after a rebase.

Human architect: "I've reviewed the design in #42 against the lens. Approved."
You: "Then run /approve-architecture 42 to record it."
Human architect: /approve-architecture 42   ← CORRECT: a human invokes it. The
                                              skill is human-only (#1042), so
                                              the model cannot.

Relationship to other approval skills

Skill	Marker (repo-qualified, see AgDR-0060)	Gate hook	Who invokes
/approve-merge	<owner>__<repo>__<pr>-ceo.approved	block-unreviewed-merge.sh	On explicit CEO per-PR merge nod
/approve-design	<owner>__<repo>__<pr>-design.approved	require-design-review-for-ui.sh	On explicit designer per-PR design nod
/approve-architecture	<owner>__<repo>__<pr>-architecture.approved	require-architecture-review.sh	On explicit architect per-PR design-review nod (or Tariq writes it on APPROVED)

All follow the same pattern: verify PR state → verify Rex marker → write marker at ops fork root → confirm → stop. None runs gh pr merge.

---

Part of ApexYard — multi-project SDLC framework for Claude Code · MIT.