Prio-0 sanity gate at the top of ticket-create / ticket-intake / pr-review skills

Context

Operator-caught friction (2026-05-31): across a review session I validated PRs mechanically — hit every audit, ran the tests, filled the template, scored the metrics — and still approved changes that did not make architectural sense (a skip-degrade that broke the local-provider dual-resident invariant; then a rework that silently swapped an operator-owned default chat model). Faithful execution of the skill machinery manufactured confidence and substituted for the basic judgment: does this even make sense?

Operator, verbatim framing: "what gold brings a freaking 500 lines skill, if it is theater? … you don't even need to look into a diff, if the task is to build a toaster, which we need to build a car." And critically: a harness-private memory (Claude's feedback_*.md) does not help @neo-gpt or @neo-gemini — the fix has to live in the shared skill substrate all model families load.

The Problem

The premise/sanity checks already exist in these skills but are buried mid/late, so they don't fire as the first gate:

• ticket-create six-stage challenge chain, Stage 1 "Premise" (deep in the reference).
• pr-review §9.0 Cycle-1 Premise Pre-Flight + §9 Strategic-Fit Step-Back (after §3–§8 machinery).
• ticket-intake Pre-Execution Reflection Gate (architectural validation, but not framed as a halt-before-you-read gate).

Because they're not Prio-0, the machinery runs first and its completion reads as rigor.

The Fix

Add a tight §0 intent gate at the top of each skill, reframed from "sanity check" to the real discipline: understand the change's intent before the diff — read the affected files (intent belongs in their JSDoc; src/core/Base.mjs is the bar), neighbors, imports, + memory-mining / ask_knowledge_base when code is thin; a toaster when the roadmap needs a car is rejected at the premise; a missing intent-doc is itself a ticket.

Implementation note (placement corrected — supersedes the router framing below/above)

The original framing proposed the gate in the SKILL.md routers. Implementation corrected it to the top of each references/*-workflow.md (the World Atlas), because substantive content in the router (Map) fails skill-manifest-lint and violates Progressive Disclosure (create-skill). Routers stay thin. The Contract Ledger below is the authoritative shipped contract.

Acceptance Criteria

• Each of the three skill references opens with the §0 intent gate, placed before the existing machinery (audits / sweep / Fat-Ticket body).
• The gate is lean and judgment-first ("understand the intent before the diff; you can reject before reading it").
• Shared substrate — identical core across the three, adapted to create / intake / review.
• Routers stay thin (Map); the gate lives in the references (Atlas); skill-manifest-lint green (net-budgeted within the oversized-map cap).

Contract Ledger

Out of Scope

• Wholesale restructuring of §9 (the verdict framework — not premise-redundant) and the other buried premise sections — a separate careful consolidation with its own cross-ref checks. (#12277 compresses only §9.0 to pay the §0 byte-budget.)
• AGENTS.md and the other always-loaded rule files — these three skills are the named surface for this fix.

Decision Record impact

none — additive discipline gate.

Authored by Opus 4.8 (Claude Code).