Sync mattpocock skills and stop formatting vendored skills

[kmcginnes]

Description

Updates the vendored mattpocock/skills to upstream HEAD — the first sync since #1913 — and fixes the formatter churn that was making these syncs hard to review.

• Exclude vendored skills from oxfmt. .agents/skills/ holds skills vendored verbatim from upstream; .claude/skills/ .codex/skills/ .kiro/skills/ link to them. Formatting them rewrote content we don't own, so every sync diff filled with our own reformatting (emphasis markers, list spacing, code-fence normalization) burying the real changes. The ignore is scoped to the vendored skills/ subtree rather than the whole harness directory, so first-party agent config (e.g. .kiro/settings/) is still formatted. Doing this first means the sync commit is purely upstream — 11 of the 22 initially-touched files turned out to be pure formatter churn with zero upstream change.
• Sync the 21 vendored skills via pnpx skills update -p. Content is now byte-identical to upstream. Substantive changes:
  • wayfinder — "investigation tickets" become decision tickets; research tickets resolve via parallel `/research` subagents on throwaway branches, and are exempt from the one-ticket-per-session rule
  • prototype — prototypes are captured as a primary source on a throwaway branch rather than deleted; main keeps only the validated decision
  • to-tickets — local output is one file per ticket under `.scratch//issues/` instead of a single root `tickets.md`
  • setup-matt-pocock-skills — leads with recommended answers; skips the triage-label and monorepo sections when they don't apply
  • grilling — broadened from plans to any decision or idea; facts come from the environment, not just the codebase
  • improve-codebase-architecture — scope the scan to git-log hot spots first (YAGNI)
  • ask-matt — routing text follows the above
  • Also vendors the new per-skill agents/openai.yaml files (Codex UI metadata plus the implicit-invocation policy mirroring disable-model-invocation)
• Add the resolving-merge-conflicts skill, new upstream since the last sync.

Note for reviewers

docs/agents/issue-tracker.md is deliberately left alone. Upstream now defaults its PR-as-request-surface flag off, but ours is on — a deliberate choice from #1863 with a documented `gh pr` triage workflow. That's an upstream default change for fresh setups, not a correction to us.

On the formatter ignore: scoping to skills/ rather than the whole harness dir follows the vendored-code principle (ignore the subtree you didn't author). There's no documented recommendation from oxc or Anthropic either way; the trade-off is that a harness later vendoring content outside skills/ would silently start getting reformatted.

Validation

• `pnpm checks` passes — 0 lint warnings/errors, formatting correct across 926 files, typecheck clean across all 3 packages
• `pnpm test` passes — 2278 tests across 198 files
• Verified the vendored tree is byte-for-byte identical to upstream mattpocock/skills HEAD
• Verified oxfmt honours `ignorePatterns` for explicit file paths too, so the lint-staged pre-commit hook no longer rewrites vendored files
• Verified all 22 skills are symlinked in both `.claude/skills/` and `.kiro/skills/`, with no missing, extra, or broken links
• Confirmed first-party agent config (`.kiro/settings/lsp.json`) is still formatted under the narrowed ignore

Docs-only change — no runtime code touched.

Related Issues

• Follows Sync mattpocock agent skills to latest upstream #1913

Check List

• I confirm that my contribution is made under the terms of the Apache 2.0 license.
• I have verified `pnpm checks` passes with no errors.
• I have verified `pnpm test` passes with no failures.
• I have covered new added functionality with unit tests if necessary.
• I have updated documentation if necessary.