[Grivn]

摘要 / Summary

技能草案要求返回完整 SKILL.md，原先与复盘和评测共用写死的 16,384 token 预算，遇到 max-tokens 就整轮失败。新增只用于草案的首次 32,768 / 重试 65,536 显式预算；仅真实截断且第二预算更大时以新任务重新生成一次，与原有内容修正共用最多两次尝试，不保存截断内容，取消不重试。

Drafts previously failed before the file-validation retry because every skill operation shared a fixed 16k output budget. Add validated, draft-only initial and retry budgets (skills.draftMaxTokens / skills.draftRetryMaxTokens, integers 128–131072). The retry never silently exceeds a configured cap. Other skill operations retain their budgets. Errors retain actual attempted budgets, source UTF-8 byte size, settings location and provider-limit guidance within the UI's 500-character error field.

关联 Issue 或背景 / Related Issue or Context

Closes #570.

涉及区域 / Affected Areas

• Host 激活、Headless 或 bundle / Host activation, Headless, or bundle
• 运行时记忆 / Runtime Memory
• 项目档案 / Project Documents
• 记忆空间或 Provider / Memory Spaces or Providers
• 子 Agent 或 Agent 工作流 / Subagent or Agent workflow
• Web UI 或对话交互 / Web UI or conversation interaction
• 设置、存储或安全 / Settings, storage, or security
• CLI、RPC、命令或工具 / CLI, RPC, commands, or tools
• 安装、更新或发布 / Installation, update, or release
• 测试、构建或文档 / Tests, build, or documentation
• 其他（请说明）/ Other (explain below)

PR 类型 / PR Type

• 面向用户的功能或行为变更 / User-facing feature or behavior change
• Bug 修复 / Bug fix
• 增强或优化 / Enhancement or optimization
• 兼容性适配 / Compatibility change
• 维护或重构 / Maintenance or refactor
• 测试或构建 / Tests or build

最新代码确认 / Latest Codebase Confirmation

• 我已基于最新 main 分支开发，或在提交前已 rebase 或合并最新 main。 / I developed from the latest main, or rebased or merged the latest main before submitting.

同步命令 / Sync command:

git fetch origin main
git merge --no-ff --no-commit origin/main

Merged latest main cd1fe02c2f4ecd46a46d53c0b336e58add4821eb, including #566 and #567, while preserving both histories. The only conflicts were adjacent imports and independent configuration tests; both sides are retained. / 已通过普通合并纳入最新 main（含 #566、#567），保留全部原提交；仅相邻 import 与独立配置测试冲突，双方内容均保留。

AI 编码披露 / AI Coding Disclosure

• 完全 AI 编码：全部编程改动由 AI 产出，并由贡献者接受和审查。 / Fully AI-coded: AI produced all programming changes, which the contributor accepted and reviewed.
• 部分 AI 辅助：AI 帮助编写或修改了部分内容。 / Partially AI-assisted: AI helped write or modify part of the change.
• 未使用 AI 编码辅助。 / No AI coding assistance was used.

使用的 AI 模型 / AI model used: GPT-6 (Codex)

使用的编码 Agent 工具 / Coding Agent tool used: OpenAI Codex desktop; implementation and independent code review used separate Codex agents. / 实现与独立代码审查由独立 Codex agents 执行。

仓库规范检查 / Repository Rules

• 未修改 DSH 官方源码，未让 tsconfig 指向 DSH 源码 checkout，仅使用正式的 @deepseek-ai/* NPM 契约。 / I did not modify DSH source or point tsconfig at a DSH source checkout, and used only published @deepseek-ai/* NPM contracts.
• Client 与 Host 边界仍以浏览器安全的 src/host/protocol.ts 为准，没有在两侧重复定义 wire DTO。 / The Client and Host boundary still uses browser-safe src/host/protocol.ts as the source for wire DTOs.
• 持久化格式、RPC 权限、路径或凭据处理的变更包含兼容或拒绝路径、安全分析和相应测试。 / Changes to persistence formats, RPC authority, paths, or credentials include compatibility or rejection paths, security analysis, and tests.
• 没有提交 token、密钥、私有记忆、未脱敏日志或生成的 lib/ 文件。 / I did not commit tokens, credentials, private memory, unredacted logs, or generated lib/ files.
• 用户可见文案和长期文档已同步维护中文与英文版本，命令、配置键和路径保持一致。 / User-facing copy and long-lived documentation are synchronized in Chinese and English, with matching commands, configuration keys, and paths.
• 会改变发布制品或其元数据的 PR 已添加 changeset；仅测试、CI 或站点文档变更可不添加。 / A PR that changes a published artifact or its metadata includes a changeset; test-only, CI-only, and site-documentation-only changes may omit one.
• 新增和修改的代码、注释、文档、提交信息不含 emoji。 / New and modified code, comments, documentation, and commits contain no emoji.

兼容性与数据安全 / Compatibility and Data Safety

新增两项可选 Skills 设置，旧配置使用默认值，不迁移技能或 ledger 格式，不更改 RPC 权限、路径或凭据。生成只保存完整且通过原有校验的草案，已安装技能仍由用户明确采纳后修改。截断失败不会写入新草案，已有草案保留；关联建议回到 pending 并记录错误。取消后即使模型已返回完整结果也不保存。重试预算等于或小于首次预算可关闭截断重试；两项预算均须符合模型提供方的上限。

The new optional settings default for existing installations. No storage-format, authority, credential or installed-skill changes are introduced. Drafts still pass schema/content checks and require explicit adoption. Both attempts stay within their explicit budgets; providers with lower caps need lower settings. A model that cannot fit the full file needs a shorter skill or a higher-capacity model. All verification uses synthetic local data; the issue reporter's real gateway and Windows environment were not exercised.

本地验证 / Local Validation

执行的命令 / Commands run:

Node 24.22.2:

pnpm run verify
pnpm run typecheck
pnpm exec vitest run --dir tests tests/subagent.spec.ts tests/skill-evolution.spec.ts tests/config.spec.ts tests/task-agent-session-host.spec.ts tests/activation-agent-compat.spec.ts tests/subagent-start.spec.ts tests/lifecycle.spec.ts tests/review-tools.spec.ts
DSH_HOST_ROOT=/path/to/installed/dsh-0.2.2-rc.1 pnpm exec vitest run --config scripts/installed-dsh.config.ts tests/subagent.spec.ts tests/skill-evolution.spec.ts tests/config.spec.ts tests/task-agent-session-host.spec.ts tests/activation-agent-compat.spec.ts tests/subagent-start.spec.ts tests/lifecycle.spec.ts tests/review-tools.spec.ts
git diff --check

结果摘要 / Result summary:

• Final merged revision bc85d88a0962ce2b13617bf98696434ba5c2a846: complete verify passed, with 2,550 root tests passed / 8 skipped (170 passed files / 2 skipped), and 759 workspace tests passed / 3 skipped across 18 projects, including 95/95 Skills Source tests. Documentation, type checks, deterministic builds, Headless activation, public entry checks and package lint all passed. / 含 fix(skills): remove dead suggestion links from evidence #566、fix(host): preserve DSH task agents and allow user continuation #567 的最终合并提交已完整验证通过。
• After integrating fix(host): preserve DSH task agents and allow user continuation #567, the affected coordinator, lifecycle, task continuation, settings and skill suites passed on both the repository dependencies and official DSH 0.2.2-rc.1: 476 passed / 1 skipped across 12 files. fix(skills): remove dead suggestion links from evidence #566 was then merged without conflicts and the full verification above was rerun. / fix(host): preserve DSH task agents and allow user continuation #567 合入后，默认依赖及官方 rc.1 的受影响测试均通过；fix(skills): remove dead suggestion links from evidence #566 无冲突合入后重新执行了上述全量验证。
• The 53 skill-evolution tests include fresh-task truncation recovery, exhausted retry with a byte-identical saved ledger, invalid file rejection, disabled retry, both cancellation timing branches and no third attempt after file-validation repair. Other regressions cover unchanged non-draft budgets, persistent settings validation and actionable errors after long provider diagnostics. / 回归覆盖预算、重试、取消、持久化与错误提示边界。
• Independent review found no Critical/Important issue; the draft-budget implementation is unchanged by the integration merges. / 独立审查无 Critical/Important 问题，整合 main 未改写草案预算实现。
• No paid model calls or live-gateway reproduction. / 未运行付费模型或真实网关复现。

用户可见变更证据 / Local Feature Evidence

证据 / Evidence:

English evidence · 中文证据

Real isolated rc.1 WebUI: with self-evolution off, defaults are 32768 / 65536; saving 49152 / 98304 shows success, and both values survive page reload. Screenshots contain only the settings dialog, without private data, credentials or local paths. / 真实隔离 rc.1 WebUI 已验证默认值、保存成功和刷新后回读；截图无私有数据、凭据和本地路径。