Ruofeng Yang
5d6e40a66b
Health-audit batch (all items verified against the tree before fixing):
- docs/MODEL_COMBINATIONS{,_CN}.md: all 24 guide links were repo-root-relative
inside docs/ (docs/X.md → resolves to docs/docs/X.md); stripped the prefix.
docs/MiniMax-GLM-Configuration.md: 2 links pointed at nonexistent
docs/README.md; now point at MODEL_COMBINATIONS_CN.md where the anchors live.
skills/research-pipeline/SKILL.md: resumable-runs link missing ../.
- skills-codex-claude-review overlay: research-refine (5 spots) + paper-figure
(1 spot) prose still said the reviewer is GPT-5.4 — in an overlay whose whole
point is Claude-as-reviewer via claude-review MCP (the invocation blocks were
already correct; only prose was stale). Now says "the Claude reviewer".
- docs/ARIS_INTRO.md: "GPT-5.5 / 5.5" blind-replacement leftovers (EN+CN spots);
docs/ARIS_INTRO.html: 11 stale GPT-5.4 refs synced to the .md;
hero SVGs (aris_logo, hero_idea, hero_interaction, hero_combined): reviewer
label "GPT-5.4 xhigh" → "GPT-5.5 xhigh" (same char width, no layout shift).
- READMEs' community-skills rows: /paper-poster now marked
"(deprecated → paper-poster-html)"; skills-codex-gemini-review/paper-slides
still cited retired /paper-poster where mainline+mirror say /paper-poster-html.
- skills/comm-lit-review/SKILL.md: frontmatter name "comm-lit-review-claude-single"
→ aligned to directory name (old name referenced nowhere; mirror already used
the dir name).
- Personal-info hygiene: interview-cheatsheet's personal-info banlist (mainline +
codex mirror, 2 spots each) listed the owner's actual lab/degree/server strings
— the banlist itself was the leak; now defines CATEGORIES only. A tutorial
review.json re-disclosed the redacted strings in its fix note; genericized.
- README.md spotlight: Anti-Autoresearch counts updated to v0.5
(46 integrity across families A–H + 13 AI-style impressions + 2 advisory = 61).
Known-remaining (deliberately out of scope): CN README lags EN by ~9 News
entries + the Anti-Autoresearch spotlight (translation batch); GPT-5.4 mentions
in kill-argument/resubmit-pipeline/slides-polish and dated News entries are
deliberate legacy options/history, untouched.
Verification: link checker 0 bad across the three docs; inventory consistent;
full suite 407 passed / 16 skipped.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
53 KiB
53 KiB
