Retoken the native-surface presence gate so descriptions carry no second person

[kyle-sexton]

Context

Anthropic's skill-authoring best-practices page keeps first and second person out of a skill description because the text is injected into the system prompt. #4108 rewrote every description clause that addressed the reader, with one exception it could not take: the native-surface presence gate.

The native-overlap registry self-check (plugins/claude-ops/skills/audit-native-overlap/scripts/overlap.py, constant GATE_TOKEN) matches the literal token resolves in your session against the description of every store row whose baked.description_phrase is true, and the reverse check fails any description that carries the token without a row. Today two descriptions carry it (claude-ops audit-install-state and audit-skill-visibility), and the same token appears in ten more skill bodies, the seam-phrasing convention (docs/conventions/seam-phrasing/README.md), the audit-native-overlap SKILL.md, the store's row evidence, and the overlap tests.

Ask

Change the token to a form that names the session without addressing the reader (for example resolves in the session) in one coordinated change:

• GATE_TOKEN in overlap.py and every fixture or assertion in test_overlap.py that spells the token.
• The two baked descriptions and the ten skill bodies that carry the phrase (grep -rl "resolves in your session" plugins/*/skills/*/SKILL.md).
• The seam-phrasing convention, the audit-native-overlap SKILL.md, and the evidence strings in docs/native-surfaces/records.json that quote the token.
• A patch bump and CHANGELOG entry for every touched plugin, and a claude-ops entry naming the token change.

Verification: scripts/validate-plugins.sh reports the self-check as ok or degraded (never broken), scripts/check-changed-skills.sh origin/main shows no dropped trigger phrase, and a scan of every description for first and second person outside quoted trigger phrases returns zero hits.

Related

• fix(playbooks,skills): keep first and second person out of description prose across 36 skills #4108 (the voice sweep that keeps the token in place for now)
• feat(skill-quality,playbooks): apply Anthropic's skill-authoring best practices as checks, guidance, and recorded deviations #4069, fix(skill-quality,playbooks): correct the reserved-word basis and severity, record the load probes, anchor every record #4072 (the skill-authoring digest this follows)