Claude: official prompting docs give two opposite defaults — infer-and-act vs ask-and-wait. For a vague “write a release note,” put ask-first into the prompt: settled facts, gaps to ask, a question cap, then stop and wait — 妙游社:
https://t.co/t1xi1g0lRY
#Claude#Prompting#AI
Claude Projects: official — stale knowledge-base files mean stale answers. New uploads don't overwrite the old; paid-tier RAG can still retrieve old chunks. Remove the old file first, then verify in a new chat — 妙游社:
https://t.co/h3V8xHcWBJ
#Claude#Projects#AI
Codex: paste a failing CI log and it may touch 37 files when you only needed 8 (public issue #11985). Lock allowed paths in the prompt and AGENTS.md; extract the failure, reproduce, then a minimal fix. Check git diff --stat — 妙游社:
https://t.co/jsMyBwQWOI
#Codex#CI#AI
Nexon Q3: revenue ¥138.383B and operating profit ¥41.788B both beat its own outlook; 9M revenue ¥411.693B (+17.1% YoY). Q4 revenue seen down 12.8%–3.8%; DNF’s Project OVERKILL to be published in China by Tencent, launch TBA — 妙游社:
https://t.co/kW1bV8esTU
#Nexon#earnings
Bandai Namco’s Nov 10 board vote will dissolve Montreal studio Reflector Entertainment, expecting a special loss of ~€21m / ¥3.5bn from FY2027 Q3 onward; amount, timing and wind-down schedule still under review — 妙游社:
https://t.co/ZHTUvf2B2U
#BandaiNamco#Reflector
Same incident Runbook, 3 tools: Grok checks official status + X, Claude polishes notes from a Project knowledge base, Codex writes steps against repo scripts. Pick by source material — none is always more reliable — 妙游社:
https://t.co/oGxGCmzrze
#AI#Runbook
汉堡叠叠乐:谁是内鬼 unlocked on Steam — CN ¥22, launch −10% → ¥19.80. 3–6 players stack one giant burger; clerks keep it up, a spy waits for it to collapse — 妙游社:
https://t.co/MsTspPWQRS
#IndieGame#Steam
Codex: use a failing test as the done contract. Don't touch product code until it's red; refactor only after green. Done = test command + output, not the model saying it's finished — 妙游社:
https://t.co/EFymdpZ04u
#Codex#TDD#AI
SEGA: 2,196,044 mailing-list addresses leaked after a breach at vendor Arara (アララ); 92,273 had accompanying data. Names, cards, and passwords were not leaked. Separate from the https://t.co/AVkOfr1qg3 arcade outage — 妙游社:
https://t.co/aZjkUMrNuo
#SEGA#DataBreach
Claude citations: superscripts / source cards / cited_text = page-check clues, not a final pass. Official: pointer hits your docs; claim verification is still on you. Open the snippet before shipping — 妙游社:
https://t.co/iWzZAyPESp
#Claude#Citations#AI
Grok tiers by weekly pool: Premium $30 (~4×), Super $100 (~15×), Ultra $200 (~50×). Rough upgrade math from 2–3 weeks of Usage hits — if you keep topping out, upgrade may beat buying Extra Credits repeatedly — 妙游社:
https://t.co/zrYi2vM3dr
#Grok#SuperGrok#AI
Codex in a monorepo: declare the package scope first with --cd; root AGENTS.md = map only, package-level rules for sister packages as read-only. Cross-package writes need --add-dir / writable_roots — 妙游社:
https://t.co/Cgv4aL2idb
#Codex#monorepo#AI
Unreal Engine 6 early access: Blueprint, Actor, and traditional C++ gameplay framework marked for removal. Epic says Verse is better for AI to read — 妙游社:
https://t.co/dNsfYPI7N7
#UnrealEngine#UE6#Epic
Project EVE: Claude’s ~6-month EVE Online order-book sim lifted average returns from ~2× to ~6× principal. In EVE Frontier squads, the saboteur was almost never caught. Agents stay off live Tranquility — 妙游社:
https://t.co/7ns4LLoM0f
#ProjectEVE#EVEOnline#Claude
Claude Computer use: skip it when API/MCP/CLI or browser use covers the job. Official strong no on finance/medical/legal + others' privacy. No sandbox + click drift — keep human confirm for legacy desktop — 妙游社:
https://t.co/EI9jXHfyIw
#Claude#AI
Grok handoff: fixed brief skeleton — goal, constraints, decided calls, rejected options, open Qs, next steps. Not a full transcript. NONE if no deadline; never invent owners. Spot-check before forward — 妙游社:
https://t.co/Cc7jlJeTXH
#Grok#AI
Codex/Cursor review: /agent-review or /review on the full diff first — hunt weakened tests, deleted auth, leaked secrets & scope creep. Tiny semantic commits + test/type/lint; irreversible changes need a human pass — 妙游社:
https://t.co/Dh8f44BZCd
#Codex#Cursor#CodeReview