Add HarnessHub (Reference Harness Implementations)

[spikesubingrui-design]

Adds HarnessHub under Reference Harness Implementations.

HarnessHub is a zero-dependency Node CLI that treats frontier-model harnesses as versioned artifacts: it pulls system prompts + tool-loops from multiple upstreams, normalizes them into one canonical JSON schema, diffs them across versions/models, watches upstreams for drift, and one-click-applies a harness into your own agent via AGENTS.md.

• Repo: https://github.com/spikesubingrui-design/harnesshub (MIT, public, zero deps)
• Edited only data/projects.yaml; ran render_readme.py and verify_catalog.py ("Verification passed"). I left sync_github_metadata.py out so the diff stays scoped to the new entry — happy to run it if you prefer.

On provenance (since this indexes extracted prompts): HarnessHub stores provenance, not copies — verbatim third-party prompts are fetched at runtime and git-ignored; the tooling is MIT and unaffiliated with any vendor, and every entry carries a takedown id.

[spikesubingrui-design]

Rebased onto the latest main to clear the conflicts — re-applied just the data/projects.yaml entry and re-ran render_readme.py + verify_catalog.py (verification passed). It's a clean single-entry diff again. Thanks!

[spikesubingrui-design]

Heads up on the recurring conflicts: README.md/README_zh.md are regenerated by the daily Add agent harness catalog entries commit, so committing the rendered files made this PR re-conflict every day.

I've scoped this PR down to only the data/projects.yaml entry (the source of truth) — it auto-merges cleanly and won't keep conflicting. The entry passes verify_catalog.py's validation; your daily render_readme.py will sync the READMEs on the next refresh / on merge. Happy to also commit the rendered files if you'd prefer the full set. Thanks!