Feature request: add coordinator.domains: frontmatter to claude-api SKILL.md (for spec-coordinator tag-routing)

## Context

A separate skills framework — [Agent Upskill's spec-coordinator](https://github.com/agentupskill/skills-core/tree/main/spec-coordinator) — uses an SME registry to route cross-domain spec requests to relevant skills. The registry has two routing layers:

1. **Tag-match (deterministic)** — picks skills whose `coordinator.domains: [...]` frontmatter overlaps query tokens (confidence 1.0).
2. **Sonnet classifier (fallback)** — scores skills against the query (threshold 0.6 for picks).

The `claude-api` skill in this repo is a frequent SME on cross-domain specs (any query mentioning prompt caching, structured output, sub-agent variance, model migration between Claude versions, etc.). Today it ships without a `coordinator:` overlay so spec-coordinator picks it classifier-only.

This works (the classifier consistently scores `claude-api` above pick_threshold on relevant queries — see [agentupskill/skills-core#56](https://github.com/agentupskill/skills-core/pull/56) for the registry-discovery clarification ensuring classifier coverage on plugin-shape skills). But adding tag-match would:

- Reduce one Sonnet classifier call per spec query when token-overlap is sufficient.
- Eliminate classifier-confidence variance (±0.1 per `_threshold-calibration.md` F4.5) on queries with unambiguous Claude-API match.
- Make routing more auditable (`method: tag` with a domain-list in the routing trace vs `method: classifier` with a confidence float).

## Specific request

Add a `coordinator.domains: [...]` block to the frontmatter of `claude-api/SKILL.md`. Suggested domains (calibrated against Agent Upskill's [_routing-priors.md](https://github.com/agentupskill/skills-core/blob/main/spec-coordinator/references/_routing-priors.md) Z17 substrate-epic):

```yaml
---
name: claude-api
description: ...
coordinator:
domains:
- claude-api
- anthropic-sdk
- prompt-caching
- structured-output
- tool-use
- thinking
- compaction
- sub-agent-variance
- model-migration
- sonnet
- opus
- haiku
pickable: true # default; only set false for invoker / product-owner skills
---
```

The `coordinator:` block is namespaced under a custom-coordinator key so it doesn't conflict with existing Anthropic-defined frontmatter. Schema documented in [`_routing-algorithm.md` §Inputs](https://github.com/agentupskill/skills-core/blob/main/spec-coordinator/references/_routing-algorithm.md).

## Why upstream rather than a downstream patch

`claude-api`'s SKILL.md ships with the Anthropic skills distribution and is not editable downstream. Without the overlay shipped here, every spec-coordinator installation falls back to classifier-only routing for `claude-api`.

## Discoverability concern flagged separately

In the current Agent Upskill installation (`~/.claude/`), the `claude-api` skill surfaces in active sessions WITHOUT a plugin-namespace prefix (no `anthropic-skills:claude-api`, no `engineering:claude-api`, etc. — just `claude-api`). Other skills from this repo (e.g., `mcp-builder`, `pdf`, `xlsx`) DO surface with the `anthropic-skills:` prefix. The provenance of the unprefixed `claude-api` registration is unclear — it might be a default-bundled skill registered by the desktop app rather than installed via this repo's plugin distribution. If that's the case, the upstream fix may need to land in the bundled-skill source (the Claude desktop app or claude-code CLI) rather than here. Pointer welcome.

## Empirical context

PR-#320 ([Boogles-Playhouse/claude-skills#320](https://github.com/Boogles-Playhouse/claude-skills/pull/320)) drafted an 8-PR substrate-epic spec that needed claude-api in the panel for prompt caching + structured output guidance. The classifier sweep correctly picked it post-fix (see [agentupskill/skills-core#56](https://github.com/agentupskill/skills-core/pull/56)), but tag-match would have been faster + more deterministic. F-CAL-6 in [_threshold-calibration.md](https://github.com/agentupskill/skills-core/blob/main/spec-coordinator/references/_threshold-calibration.md) records the calibration trace.

## Acceptance

- `claude-api/SKILL.md` gets a `coordinator.domains: [...]` block.
- A short note in the marketplace README (or a new CONVENTIONS.md) explaining the field's consumer (Agent Upskill spec-coordinator) so future skill authors populate it.
- The provenance question above gets a pointer (this repo, the desktop app, or the claude-code CLI bundles).
- No breaking changes to existing consumption — the field is additive.

Happy to send a PR if Anthropic prefers that route.

## Cross-references

- Routing algorithm: [_routing-algorithm.md](https://github.com/agentupskill/skills-core/blob/main/spec-coordinator/references/_routing-algorithm.md)
- claude-api SME inventory entry (with provenance note): [_routing-priors.md "Other (registered, source uncertain)"](https://github.com/agentupskill/skills-core/blob/main/spec-coordinator/references/_routing-priors.md)
- Companion issue (cloudflare:* skills): [cloudflare/skills#49](https://github.com/cloudflare/skills/issues/49)
- Companion issue (engineering:* skills): filed against anthropics/knowledge-work-plugins (link to follow once filed)