feat: WINGMAN_MODE — review intensity staged by project maturity

[ashbrener]

Why

Wingman is already non-blocking at the hook level — but the loop it prescribes has one intensity for every project: gather, categorize, fix, converge. Applied to a young codebase that kills traction. An exhaustive review tail buys correctness nobody depends on yet, at the cost of actually shipping.

Real data from a project dogfooding this pack: 14 review rounds on one spec, 2–3 on every routine version bump. The first round or two caught genuinely dangerous defects (arbitrary code execution on adopt, a repair that deleted operator content). Rounds 8–14 were refinements of refinements in an optional, default-disabled feature.

What this adds

WINGMAN_MODE — env var → .wingman-mode repo file → default advisory:

Mode	When	Behavior
off	Early — pre-users, iterating fast	Skip review entirely. Say so once per session, never nag.
advisory (default)	Maturing — real consumers, shipping	One round per PR. Fix what's cheap and obviously right, record the rest, move on.
strict	Mature — artifacts that propagate to other repos, destructive paths, security surfaces	Round per push, converging while findings keep landing. Still advisory.

What this makes explicit

• Review never blocks a merge and never blocks a push. What blocks is the project's own lint/typecheck/test gate — fast, deterministic, about things that actually break.
• Findings not fixed are recorded, not owed. A deferral is an audit trail with a one-line rationale; follow-ups land in follow-up PRs, not before the merge.
• Three classes still deserve a look in any mode: destructive paths, artifacts the repo stamps into other repos (a defect there propagates and needs a re-stamp to retract), and credential/.env handling.

Docs-and-skill scope: the hook already runs in the background and already never blocks. off short-circuits the auto-surface protocol; the deeper plumbing (hook honoring off before invoking the reviewer) is a natural follow-up.