Skill-Body Wiring + Graph-First Expansion

Goal

Wire the seam worklist into /impl-hygiene-review Phase 2 (landscape produces the worklist) and Phase 3 (analysis consumes it + runs the verification gate). Expand the thin Graph-First Mandate pointer into the real per-category query map. Keep the skill body script-first compliant (queries live in the runtime; the body cites payload) and @-include compliant.

Implementation Sketch

Edit SKILL.md + phase-2-landscape.md + phase-3-analysis.md so the worklist is the Phase-3 reading agenda (graph-nominated targets replace the blind dependency-expanded sweep). The skill body carries directive + citation only; no composed cypher in prose (that is a script-first violation).

Spec References

.claude/rules/script-first.md §1/§3 (orchestration in runtime; 4-axis rubric); .claude/rules/skill-template.md §2 (@-include SSOT); .claude/rules/intelligence.md Graph-First Mandate.

Work Items

• Wire Phase 2 (phase-2-landscape.md) to produce the seam worklist via the runtime
• Wire Phase 3 (phase-3-analysis.md) to consume the worklist + run the verification gate per candidate
• Expand the Graph-First Mandate section into the per-category query map (cite runtime, not inline cypher)
• Verify @-include + script-first rubric stay clean on the edited skill files

Items

• Add a Phase 0 graph-refresh step to /impl-hygiene-review: ALWAYS refresh the Ori code graph (scripts/intel-query.sh refresh —code, dual-repo, recomputes the GDS hygiene properties) before producing the seam worklist, so the scan reflects the working tree incl. uncommitted edits; graph-unavailable degrades gracefully (skip the graph tier, fall back to the file-walk)