sogeisetsu · Daily GitHub activitysogeisetsu 的 GitHub 每日动态
@sogeisetsu·2026-10-23 00:00 – 24:00 · UTC+8
58 commits were recorded across four repositories, with 37 additional automated commits by github-actions[bot].
In sogeisetsu/sopen-translate, 40 own commits redesigned the options and popup UI, added NVIDIA presets, Riva protocol support, a local debug log backend, and asynchronous review with ID-based patches and undo.
In sogeisetsu/sogeisetsu, 10 own commits added CLI help, decoupled providers into config/providers.json, connected NVIDIA to the daily-report chain, and retired the Minimax and Longcat free legs.
The sogeisetsu/opencode-skill-management repository received 7 own commits for a bilingual Astro and Starlight documentation site, Pages deployment, release policy, and version preparation. sogeisetsu/opencode-agent-sandbox added one own commit documenting planning capabilities. sogeisetsu/opencode-skill-management published v1.0.0, introducing a stable skill switchboard with list, inspect, disable, and restore operations that modify only a global permission rule.
Separately, yulia-ivashko closed openchamber/openchamber issue #5489 after openchamber-bot[bot] confirmed that the Improve action armed an implement objective unconditionally.
当天共记录 58 次提交，另有 37 次由 github-actions[bot] 自动提交。
sogeisetsu/sopen-translate 的 40 次本人提交重做了选项与弹窗界面，加入 NVIDIA 预设、Riva 协议、本地调试日志后端，以及基于 ID 补丁和撤销功能的异步审查。
sogeisetsu/sogeisetsu 的 10 次本人提交新增 CLI 帮助，将 provider 解耦到 config/providers.json，把 NVIDIA 接入日报链首，并下线 Minimax 与 Longcat 免费链。
sogeisetsu/opencode-skill-management 的 7 次本人提交涵盖双语 Astro 与 Starlight 文档站、Pages 部署、发布政策和版本准备。
sogeisetsu/opencode-agent-sandbox 的 1 次本人提交收录了规划类能力。
sogeisetsu/opencode-skill-management 发布 v1.0.0，提供稳定的技能开关面板，支持列出、检查、禁用和恢复技能，禁用时仅写入一条全局权限规则。
此外，openchamber-bot[bot] 确认 Improve 操作会无条件设置 implement 目标后，yulia-ivashko 关闭了 openchamber/openchamber issue #5489。
37 automated commits by @github-actions[bot] · not counted in the total37 次自动提交，由 @github-actions[bot] 发起 · 不计入总提交数
skill_management adds a switchboard for the skills installed in your OpenCode setup: see what is installed, find out why one will not load, take one out of the model's view, and put it back.
Disabling never touches skill files. It writes one skill permission rule into your global config (effect: "deny", or "ask" for a soft disable). Nothing is deleted, moved or rewritten, and removing that rule restores the skill exactly as it was.
This is the first stable release: the surface described below is frozen for 1.x.
| Action | What it does | Writes config |
|---|---|---|
list | Every registered skill with its effective state for the calling agent (available / disabled / ask) | no |
inspect | Why one skill is (un)available: the matching rule, per-agent differences, how its id was derived | no |
disable | Hide it from the model and reject loading it (effect: "ask" keeps it listed and asks on load) | yes |
enable | Remove the blocking rule again, whoever wrote it | yes |
How a permission is resolved: the host reads the rule set of the agent that is calling, takes the last match (findLast) with * / ? globs, and — when nothing matches — falls back to ask, not allow. That is why list and inspect evaluate per agent and name the agent they used.
disable writes exactly this line (the trailing comm…
AI This first stable release adds a switchboard for OpenCode skills, supporting list, inspect, disable, and restore operations. Disabling writes one skill permission rule with effect: "deny" or "ask" and does not delete, move, or rewrite skill files; the described surface is frozen for 1.x.首个稳定版本提供 OpenCode 技能开关面板，支持列出、检查、禁用和恢复。禁用操作仅写入一条 skill 权限规则，effect 为 "deny" 或 "ask"，不会删除、移动或重写技能文件；所述功能在 1.x 版本冻结。
For the maintainer: fix-ready — cause traced Your analysis is right; I verified each link in it against current main, and the arm is unconditional for the improve action. `handleConfirmPlanSe…