sogeisetsu · Daily GitHub activitysogeisetsu 的 GitHub 每日动态
@sogeisetsu·2026-10-18 00:00 – 24:00 · UTC+8
sogeisetsu logged 14 own commits across three repos, plus 28 automated commits by github-actions[bot].
Seven went to sogeisetsu/opencode-skill-management: implementing the skill_management plugin (list/inspect/disable/enable), moving plans to .openchamber/plans and switching line endings to LF.
In sogeisetsu/sogeisetsu six own commits refreshed AGENTS.md, stopped duplicate creation of the resident issue, narrowed the additive-phrasing invariant in verify, and reworked AI failure visibility, model labeling and the 30-day summary.
One commit in sogeisetsu/opencode-agent-sandbox added an opencode-provider knowledge category on catalog sources, credentials and endpoints.
Issue activity was 1 item: a comment on openchamber/openchamber#4939 about the rendered-page selection menu. sogeisetsu started discussion openchamber/openchamber#5114, proposing to split that fix with the frame→host selection bridge first.
Two replies arrived: btriapitsyn said #4601 was fixed in 7d037b07d, and yulia-ivashko could not reproduce #4899 at 150% or 200% scaling. btriapitsyn also closed #4896, #4920 and #4601; no pull requests, reviews or releases were recorded.
2026-10-18,sogeisetsu 在三个仓库共完成 14 次本人提交,另有 github-actions[bot] 的 28 次自动提交。
其中 7 次在 sogeisetsu/opencode-skill-management:实现 skill_management 插件(list/inspect/disable/enable),把计划移到 .openchamber/plans,并将换行符统一为 LF。
sogeisetsu/sogeisetsu 有 6 次本人提交,刷新了 AGENTS.md,修复常住 issue 被重复创建的问题,收窄 verify 中「相加表述」不变式的扫描范围,并改版 AI 失败可见性、模型标注与 30 天总结。
sogeisetsu/opencode-agent-sandbox 有 1 次提交,新增 opencode-provider 知识类别,涵盖 catalog 来源、凭据归属与端点。
issue 相关活动共 1 项:在 openchamber/openchamber#4939 下留言,讨论渲染页内选中文字无法唤出评论菜单的问题。
sogeisetsu 新开了讨论 openchamber/openchamber#5114,建议先做 frame→host 选区桥接,再拆分该修复。
收到两条回复:btriapitsyn 表示 #4601 已在 7d037b07d 修复;yulia-ivashko 表示在 150% 与 200% 缩放下无法复现 #4899。
btriapitsyn 还关闭了 #4896、#4920 和 #4601;当天没有 PR、评审或发布。
28 automated commits by @github-actions[bot] · not counted in the total28 次自动提交,由 @github-actions[bot] 发起 · 不计入总提交数
Fixed in 7d037b07d. The thinking level now stays with the model it was set for. Switching models drops it, and the new model starts on its own level: whatever you picked for it earlier in the session…
We couldn't reproduce this. On Windows 11 at 150% scaling, a Desktop (1440×900) capture came out at 2160×1350 on both current main and 2.3.0, and with the display forced to 200% it came out at 2880×1…