spec-tdd-escalatelisted

# spec-tdd-escalate **REQUIRED BACKGROUND:** You MUST understand the `spec-tdd` family first — `spec-tdd` (general), `spec-tdd-coverage` (branch-coverage evidence), `spec-tdd-adversarial` (correctness-critical: money/auth/data-loss, independent attacker), and `spec-tdd-lite` (in-session: one small non-critical unit, no delegation). This skill is the **no-grill front-end**: it only routes. ## Overview The requirement is already settled (a locked plan, spec, or doc). Your ENTIRE job is to **run the I21 pre-flight (below), read it, run the fuzziness sniff, pick the tier that fits the stakes, and invoke that tier by name.** Then stop. **Core principle: route-only.** You do NOT grill, do NOT write the acceptance test, do NOT delegate, do NOT gate. The tier you invoke runs its own Phase 1 (it writes the test) and executes from there — the delegated tiers dispatch an implementer; `spec-tdd-lite` implements in-session. Escalate adds exactly three things over calling a tier directly: **the machine picks the tier, the I21 pre-flight refuses to run this protocol on a non-top-tier orchestrator unheard, and the fuzziness sniff refuses to silently route a doc that only looks settled** — a clean doc on a top-tier session never triggers a question. > Sibling front-end: `grill-spec-tdd` = requirement FUZZY → grill + gate the spec + write the test from the final spec + route. **escalate = requirement SETTLED → route only.** If you'd need to interrogate, you're in grill-spec-tdd territory,