feat(checks): add model-unknown warn rule for the model frontmatter field

[protosphinx]

Why

Skills can specify a model: field in their frontmatter to pin the Claude
model used at runtime. A typo there (claude-sonet-4-6 instead of
claude-sonnet-4-6) causes the skill to silently run on the wrong model or
fall back to a default, with no feedback from the model output. This is
exactly the class of silent bug skillcheck is designed to catch.

This PR adds a model-unknown warn rule (not error, because the model list
evolves with each Anthropic release) that validates the model: field
against a curated set of known Claude model IDs.

Also fixes a docs omission: the existing empty-body rule was not listed
in the README "What it checks" table.

What

• New constant KNOWN_CLAUDE_MODELS in src/checks.ts covering Claude 3.x
  and 4.x series model IDs
• New check function checkModelUnknown called in the per-skill loop inside
  runChecks
• 5 new tests covering: unrecognized model, recognized model, absent field,
  message content, and a 3.x series model
• README table updated with empty-body and model-unknown rows

Tests

• test/checks.test.ts: 5 new tests for model-unknown
• All 104 existing tests continue to pass
• npm run lint && npm run build && npm test all green locally

Self-merge gate

• all CI checks pass
• LOC delta < 250 (84 lines added, 0 removed)
• no public-API surface change (does not touch src/index.ts)
• no runtime-dependency additions
• no workflow file changes
• tests added or extended

---

Generated by Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 66e4d9d01f

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Accept documented model aliases and inherit

This whitelist makes valid skill frontmatter produce model-unknown warnings: I checked the Claude Code skills frontmatter docs, which say model accepts the same values as /model, or inherit, and the CLI reference documents /model aliases such as sonnet and opus. A skill using model: inherit or model: sonnet will now be reported as a typo, and in --strict CI that false positive becomes a failing run, so the accepted set should include the documented non-ID values (or delegate validation to Claude Code's model parser).

Useful? React with 👍 / 👎.