chore(sync): record CLAUDE.md and the bench scaling as skipped

[IgorShevchik]

First two of six in the current queue. Both deliberately not taken; adjacent, so batched per PORTING.md §6 step 4b. No code changes.

1. nuxt/ui@6f82268b — add CLAUDE.md importing AGENTS.md → skip

Three lines adding a tracked CLAUDE.md that imports AGENTS.md. We have AGENTS.md (156 lines) and no CLAUDE.md, so it would apply cleanly — and would still be wrong:

# AI
CLAUDE.md
.claude/settings.local.json
.claude/channels/
.claude/debug/

CLAUDE.md is in our .gitignore at line 51. The company it keeps settles it: this fork classifies the file as per-developer Claude Code state, not as tracked instructions. Porting the commit would mean deleting that line and reversing a standing decision — not something a sync port should do on its own authority.

The trade-off is named rather than hidden: upstream's comment says Claude Code auto-loads only CLAUDE.md, so our tracked AGENTS.md is not loaded automatically, and each developer supplies their own ignored copy. If that is judged a mistake, it is a two-line change — .gitignore line 51 plus this commit — and it should be made on its own merits.

2. nuxt/ui@3d2129d2 — scale tv benchmarks past codspeed's noise floor → skip

Wraps every bench body in a 100-iteration loop, because CodSpeed runs them under an instruction-counting simulator and GitHub's runners alternate between Intel and AMD, so sub-millisecond bodies collect phantom regressions.

We have test/bench/tv.bench.ts and the change would apply cleanly. But this fork does not run CodSpeed: bench is a plain vitest bench --project vue, no workflow invokes it, and nothing compares runs across machines.

So it would buy nothing, cost a hundredfold longer local run, and add a comment describing CI we do not have. Worth taking the day benchmarks are wired into CI — and taking the comment with it.

A larger finding, surfaced by the first one

Checking why CLAUDE.md was ignored led to git log -S 'CLAUDE.md' -- .gitignore → b55bd3e7, a deps port. Its body never mentions CLAUDE.md, but it does say:

Skipped: ai + @ai-sdk/* (b24ui stays on the deferred v6 line)

Five other ledger entries say the same. That is what prompted #438: the ai v6 line was a deliberate, repeatedly reaffirmed deferral rather than drift, deferred for exactly the DeepSeek provider coupling that #425 then broke. Recorded in both logs here, and as a §2 invariant in #438.

Ledger

cursor → 3d2129d2. Both entries added with decision: skip and the reasoning in .sync/log/. .sync/dep-parity.json refreshed to the new cursor per the §6 step from #429 — neither commit touches a manifest, so only the cursor line moves and all 148 versions are identical.

Verify (CI=true)

lint and the test/utils suite (495 tests, 25 files). Documentation and ledger only; nothing executable changes.

Remaining in the queue: ae2bd5eb (Splitter, adopting), 2beb2345 + a630c943 (showcase, no-ops), 14ac2438 (ProgressGroup plus the --percent fix for the existing Progress).

---

Generated by Claude Code