claude-memory:audit — add reciprocal boundary note routing model-era instruction findings to claude-config:audit-instructions

[kyle-sexton]

Context

PR for #800 ships claude-config:audit-instructions with a one-way boundary partition: on memory-layer surfaces (CLAUDE.md, CLAUDE.local.md, .claude/rules) it runs only the model-era checks I6–I11 and routes hygiene findings (its I1–I5 seeds) to claude-memory:audit, whose C1–C8 already implement that layer. claude-memory:audit does not yet point back the other way, so a user landing there with a model-era question ("are these instructions holding the model back?") gets no route to the sibling. The note is a separate-plugin change deliberately split out of the #800 PR to keep that PR single-plugin.

Proposed work

• Add a reciprocal scope-boundary note to plugins/claude-memory/skills/audit/SKILL.md: model-era instruction-content findings (prior-model workarounds, over-prescriptive scaffolding, bare prohibitions, reasoning-echo directives, stale example scaffolding) route to claude-config:audit-instructions, seam-phrased per docs/conventions/seam-phrasing/README.md (gate "when the claude-config plugin is installed" + stated fallback).
• Patch-bump claude-memory with a CHANGELOG entry per changelog-parity.

Acceptance criteria

• claude-memory:audit SKILL.md names claude-config:audit-instructions as the route-out for model-era instruction-content findings, with install gate and fallback.
• Wording stays out of audit-instructions' trigger phrase space (capability-framed, no duplication of its check catalog).
• check-skill, validate-plugin-contracts, and changelog-parity pass.

References

• claude-config: add model-upgrade unhobbling audit — sweep instruction surfaces for constraints newer models no longer need #800 (the audit-instructions skill; D1 partition decision recorded in its PR)
• plugins/claude-config/skills/audit-instructions/SKILL.md "Scope boundary (route out)"
• docs/conventions/seam-phrasing/README.md

[kyle-sexton]