review/land dispatchers leak an ambient gateway env onto the plain claude route

[SSFSKIM]

Problem & intent

review-dispatch.sh and land-dispatch.sh have the same gateway-env leak that
PR #35 just fixed in implement-dispatch.sh: the plain (claude) spawn branch
invokes daemon-spawn.sh with no environment prefix, so an ambient
DAEMON_CLAUDE_SETTINGS / DAEMON_CLAUDE_EFFORT is inherited. daemon-spawn.sh
then applies --settings/--effort AND persists both into the registry meta,
and daemon-resume.sh restores them on every later fork — so the worker rides
the clodex gateway for its whole life while the dispatch log says
engine=claude.

Discovered by the review worker on PR #35 (surfaced by the review engine on the
implement dispatcher, generalized by the reviewer). Confirmed mechanically on the
implement lane before the fix: with DAEMON_CLAUDE_SETTINGS=/tmp/x.json DAEMON_CLAUDE_EFFORT=high exported,
bash tests/implementing-tickets/test-implement-dispatch.sh failed its
"no gateway env" assertions with
spawn-env:settings=/tmp/x.json;effort=high. The same shape is present at:

• skills/reviewing-prs/scripts/review-dispatch.sh — plain branch around the
  daemon-spawn.sh call (the gateway branch is at :407-408)
• skills/reviewing-prs/scripts/land-dispatch.sh — plain branch around the
  daemon-spawn.sh call (the gateway branch is at :313-314)

Why it matters now rather than later: the review and land lanes still default to
codex, so today only an engine:claude label or WORKER_ENGINE=claude reaches
the leaking branch — narrow but real. #32 (C5) flips those two lanes to a
claude default, which makes the leaking branch the COMMON path, exactly as
happened on the implement lane in #33/#35. Fixing it before or as part of #32
avoids shipping the same silent-misrouting default twice.

Constraints

• Overlaps C5: review-loop worker default route clodex → claude + plumbing removal #32 by file and by line: C5: review-loop worker default route clodex → claude + plumbing removal #32 rewrites these same two spawn branches.
  Whoever takes this should either fold it into C5: review-loop worker default route clodex → claude + plumbing removal #32's change or land it first and
  let C5: review-loop worker default route clodex → claude + plumbing removal #32 build on it — do not land two competing edits to the same lines.
• Do not change the review/land default route, the engine:* label precedence,
  or WORKER_ENGINE semantics here. This ticket is the env-hygiene fix only;
  the default flip is C5: review-loop worker default route clodex → claude + plumbing removal #32's.
• Match the fix shape already landed on the implement lane
  (implement-dispatch.sh, PR feat(implement): default model route flips clodex → claude (C6) #35, commit d27fbcc): the plain branch clears
  both variables with a command-prefix assignment, mirroring the gateway
  branch's existing prefix style. Note VAR= alone trips shellcheck SC1007 —
  the landed form is VAR=''.

Success criteria

• A plain-route review dispatch and a plain-route land dispatch each hand
  daemon-spawn.sh no gateway settings and no gateway effort, regardless of what
  the ambient environment exports.
• A regression assertion proves it in tests/reviewing-prs/test-review-dispatch.sh
  and tests/reviewing-prs/test-land-dispatch.sh. The implement lane's approach
  is a good precedent: export an ambient gateway route for the whole suite, which
  converts every existing "no gateway env" assertion into a leak regression and
  simultaneously strengthens the gateway-route assertions into proof that the
  prefix overrides an ambient value rather than coinciding with an empty one.
• The gateway route itself keeps working unchanged in both lanes.

Open questions

• Should this be fixed at the substrate instead of per-dispatcher? Raised by the
  PR feat(implement): default model route flips clodex → claude (C6) #35 fixer as out-of-scope scope-creep: daemon-spawn.sh gives a caller no
  way to declare "plain route" other than passing empty strings — the contract is
  env inheritance, which is what makes this leak class available to every
  caller, including future ones. An explicit --no-gateway flag or a route
  argument would make the plain route unspoofable in one place rather than in
  three. Recommend: land the three-line dispatcher hygiene fix now (it is what
  C5: review-loop worker default route clodex → claude + plumbing removal #32 needs), and treat the substrate contract change as a separate design
  decision — it touches every daemon caller and wants its own gate.

Decision log

• Registered by the PR feat(implement): default model route flips clodex → claude (C6) #35 review worker as a TOO BIG scope fork rather than
  waved into PR feat(implement): default model route flips clodex → claude (C6) #35: ticket C6: implement worker default route clodex → claude #33 explicitly fenced the review and land lanes off
  to a later unit, so fixing them inside feat(implement): default model route flips clodex → claude (C6) #35 would have been scope the ticket
  ruled out. The implement lane's own leak WAS fixed in feat(implement): default model route flips clodex → claude (C6) #35 (in scope — it was
  the line that PR changed).
• Filed as its own ticket rather than a comment on C5: review-loop worker default route clodex → claude + plumbing removal #32 because it is a distinct
  defect with its own acceptance criteria and is a live (if narrow) bug today,
  before any default flip. The file/line overlap with C5: review-loop worker default route clodex → claude + plumbing removal #32 is recorded above as a
  constraint instead.