Lane X: authoring doctrine — one-skill-per-call + invocation-reach audit

[kyle-sexton]

Parent

Refs #2933. Source: topic pocock-shipping-breakdown (PR backfilled when opened).

What to build

Three cross-cutting items from upstream's post-v1.2.3 delta (PRs #878/#880 in mattpocock/skills; evidence: topic memory slice upstream-bringover-audit.md):

• C22 invocation-reach audit (the priority): upstream found six live violations in their own repo of the invariant our SSOT already tracks — skills instructing Skill-tool invocation of a user-invoked-only (disable-model-invocation: true) target, which no skill can reach. Audit our fleet for the same defect class; adopt the canonical human-relay rewording ("tell the user to run /X") where found; consider a standing skill-quality:check / audit-instructions check. This is the SSOT tracked-strand's second trigger condition firing.
• C21 one-skill-per-call authoring line: "the Skill tool takes one skill per call; a step needing two skills is two calls" — cheap doctrine line for playbooks:skill-authoring / skill-quality:check.
• C23 concrete-artifact trigger phrasing (minor): compare domain-driven-design:curate-language triggers against upstream's artifact-anchored rewording.

Update the SSOT tracked-strand rows with the outcome.

Acceptance criteria

• Fleet audit completed with findings list; violations fixed with human-relay phrasing
• C21/C23 verdicts recorded; accepted doctrine lines landed
• SSOT invocation-reach strand updated (fired-and-resolved or narrowed)

Blocked by

None — can start immediately.