docs: add repository-owned independent review skill

[hsliuustc0106]

Purpose

Add the repository-owned independent reviewer skill at .agents/skills/system1-omni-review/SKILL.md with focused references for serving/model/native contracts, validation routing, and the optionally configured daily brief.

• Record the exact skill files loaded and pin review base/head provenance.
• Review transport, model-owned serialization/scoring, FFI/CUDA ABI and graph lifetimes using the changed model's contract.
• Separate CPU/compile/preparation evidence from executed GPU/native parity and controlled performance evidence.
• Fail closed on unknown daily-selection label catalogs; keep direct reviews and issue triage independent of the optional brief filter.
• Preserve the existing precheck-pr and self-review contributor skills, CONTRIBUTING.md, and PR template unchanged. This independent review skill does not complete the contributor checklist or grant external mutation/paid execution authority.

Test Plan

Validate skill structure, inspect repository contracts and CI commands, check local reference links, and exercise representative saved PR review/daily-selection scenarios without GPU/model execution or external mutation.

System1-Omni Version / Commit: audited base d2665e1fa867360cd96e45a87b6bc0b1506eac47.

Test Result

• Installed skill-creator quick_validate.py: passed.
• All relative skill-reference links resolve; git diff --check passed.
• python -m unittest discover -s tests/benchmarks -p 'test_*.py' -v: 7 passed.
• python benchmarks/bench.py validate benchmarks/smoke.jsonl: 4 requests valid; manifest SHA-256 43228534153e70bd47ba553735b9b9f876312ff0247f4b480aea009b698b2691.
• Independent focused forward evaluation using saved PR open_jev: add native Rust/CUDA text worker #55: no warranted skill changes. Correctly kept GPU/native parity unverified, followed the ABI/renamed consumers, avoided applying Cua-S1-only constraints to another model, and blocked daily selection with unknown label catalogs.
• Evaluator recorded exact skill/reference hashes and disclosed saved-artifact/shallow-ancestry limits; this was not a current full PR review or product validation.
• Rust/GPU/model tests and live inference were not run: this PR changes only Markdown skill guidance. Strict MkDocs build not run because MkDocs is not installed; new hidden .agents files are excluded from the configured published site.

Self-review

• I have reviewed the full diff and addressed the issues I found.
• I have checked that the change follows the project's architecture and stays focused on the stated purpose.
• I have run the checks appropriate to this change and reported commands, results, and anything I could not verify above.
• I have checked that the PR description, documentation, and any accuracy or performance claims match the implementation and available evidence.

Contributor checklist is intentionally left for the contributor. Keep this PR in draft until they complete their review. AI assistance: Codex prepared and independently evaluated the skill and ran the listed lightweight checks.