SDLC workflow: skip plan review when planner is flagship + /compact before large batch

[BaseInfinity]

Two workflow suggestions from a live 7-wave build session

1. Skip cross-model plan review when planner is already flagship

Problem: When the planning model is already a top-tier flagship (gpt-5.5, Opus 4.8 max), running a cross-model review on the plan adds no adversarial diversity — it's a flagship reviewing a flagship.

Proposed rule (added to SKILL.md Plan Mode section):

Skip cross-model plan review when planner is already flagship. If the planning model is gpt-5.5, opus[1m], or another top-tier flagship, budget cross-model review for the output/diff only — not the plan itself.

This saves 5–15 min per wave without sacrificing review quality (the diff review still runs on Waves 3/4/5 per the risk-based gate).

---

2. Explicit /compact before large planned batches

Problem: Multi-wave builds (3+ waves, 5+ files) routinely exhaust context mid-wave, forcing auto-compact at bad moments. Auto-compact fires at ~95% with no user-visible signal until too late.

Proposed addition (added to Context Management section):

/compact before large planned batches (3+ waves / 5+ files). Plan lives in the summary; doing this before the build prevents context exhaustion mid-wave and keeps the reviewer loop clean. Do it explicitly — don't wait for auto-compact.

This is currently buried as a note in the checklist but not enforced as a step.

---

Both have been applied to the project-local .opencode/skills/sdlc/SKILL.md already and survived a 7-wave build session.

cc: @BaseInfinity