Home / All Skills / Workflow & Core Tools / dbs-agent-migration
English (EN)简体中文 (ZH)日本語 (JA)
Workflow & Core Tools · dontbesilent2025/dbskill

dbs-agent-migration

Audit project rule files, identify the source of truth, standardize naming, and generate bridges to migrate projects into a multi-client consistent Agent workbench. Use when users request migrating Claude Code, Codex, Grok, general Agents, or organizing AGENTS.md.

L2 Local · Restricted workspaceV1 Commit-PinnedWrite FileDual-pass Audit
Author dontbesilent2025GitHub dontbesilent2025/dbskill ↗Stars 10121License CC BY-NC 4.0（仓库根 LICENSE 首行「Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)」，条款为 Attribution + NonCommercial，Copyright (c) 2026 dontbesilent。**非商业授权**：个人/学习/研究可直接使用，商业用途需单独授权；GitHub API license.spdx_id 为 NOASSERTION（CC 系非 SPDX 标准标识））Commit 8b8e33f1ec
Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

Migration is defined as 'structural rearrangement' rather than copying files: simultaneously resolving at least five areas: project-level rule files, skill source-of-truth location, bridge naming conventions, multi-client consistency, and sustainable maintainability.

skills/dbs-agent-migration/SKILL.md
**这不是安装教程。也不是脚本执行器。**
What is being done here: explicitly disqualifies partial migrations like 'copying CLAUDE.md to AGENTS.md' (Principle 1 notes it 'at most only solves getting things to run initially'), raising the delivery standard to the 4-part suite of 'source of truth + naming + multi-client bridge + maintainability'. This distinguishes it from ordinary installation tutorials—it purely audits, consolidates, generates, and verifies throughout, without running installation commands.

Source of truth takes priority: the in-project skills/ is the ideal source of truth, while ~/.claude|.codex|.grok|.agents/skills are merely bridges or host installation entry points; long-term logic must not be written inside bridges.

skills/dbs-agent-migration/SKILL.md
- `~/.claude/skills/`、`~/.codex/skills/`、`~/.grok/skills/`、`~/.agents/skills/` 都只是 bridge 或宿主安装入口
Accompanied by the prohibition 'do not maintain long-term logic inside bridges' and the closing maintenance guideline: 'How to maintain in the future: only modify the source of truth, then regenerate the bridge for the corresponding host.' This is the core invariant of the entire methodology: change one source of truth, and multiple clients inherit it consistently via thin bridges.

Do not assume the project is already standardized: explicitly enumerates 4 rule-layer states (complete partial migration / CLAUDE.md only / AGENTS.md only / completely scattered) × 5 host states (Claude only / Codex only / Grok only / general Agents only / multi-client but inconsistent) as branching criteria.

skills/dbs-agent-migration/SKILL.md
- **D 类**：没有规范，skill 散落
Structuring this as a matrix rather than a linear flow is the key difference between this skill and a 'one-off migration script': auditing classifies first (A/B/C/D + a single host assessment), then decides which layer to touch next; the Phase 1 output format strictly requires reporting 'which category / what was done right / what is genuinely missing / recommended layer to touch first' and asking whether to continue.

Each step in the 6-phase process closes with 'pre-write confirmation': Audit → Rule file migration → Source of truth identification/establishment → Naming and frontmatter → Multi-client bridge generation → Verification.

skills/dbs-agent-migration/SKILL.md
不要一口气做完再汇报。
Phase 2 requires explaining before writing 'whether to create or rewrite which file / what will be kept / what will be deleted / why layered this way'; Phase 4 specifies 'must provide the user with an inventory rather than directly moving files'; Phase 5 requires first explaining which hosts bridges will be generated for and obtaining confirmation. Multi-step confirmation is explicitly defined here as 'part of the product'.

Grok bridges have hard constraints and precise templates: frontmatter must include user_invocable: true, description must specify the trigger method in Grok TUI and 'read source of truth first', and the body should recommend a Grok Bridge section + absolute path to the Source of truth.

skills/dbs-agent-migration/SKILL.md
Grok bridge **必须** 在 frontmatter 里包含 `user_invocable: true`，否则用户在 Grok TUI 输入 `/` 后搜不到这个 skill。
The template skeleton is provided verbatim (multiline name / user_invocable / description + ## Grok Bridge section), along with three 'must-check' items (whether user_invocable exists, whether description mentions Grok TUI and trigger words, and whether the path is absolute). Such host-specific hard constraints concretely anchor the 'multi-client consistency' goal—shifting silent failures like 'cannot search on the Grok side' into generation-time checks.

The general Agents side prioritizes symlinks over copying and applies a conservative strategy for same-name conflicts: no overwriting, report the target path and current type, and let the user confirm whether to move out the legacy directory.

skills/dbs-agent-migration/SKILL.md
`~/.agents/skills/<name>` 优先使用软链指向真源目录。这个目录已知会被豆包 Mac App、Trae Solo 和 Codex 读取。
The same section requires: if the target is already a symlink with the same name, it can be updated, but 'after updating, readlink must be checked to ensure it points back to the expected path'; the Phase 6 verification checklist includes a dedicated item: 'whether targets in `~/.agents/skills` have real directory conflicts or dangling symlinks'. The symlink strategy avoids drift across multiple copies of the source of truth, landing the 'modify source of truth only' invariant at the filesystem layer.

Naming standardization has verifiable rules: each Skill retains only 1 invocable name in lowercase English kebab-case, formal dbskill Skills use the dbs- prefix, directory names, frontmatter name, and bridge directory names must match completely, Codex display_name must equal the standard English name, and templated short_description is prohibited.

skills/dbs-agent-migration/SKILL.md
可调用名使用小写英文 kebab-case；dbskill 正式 Skill 使用 `dbs-` 前缀，例如 `dbs-good-question`。
The same paragraph includes 6 numbered rules, including 'Chinese names may only serve as explanatory titles and natural language intents, not as `/` invocation aliases' and 'do not let scripts invent temporary names based on titles'; this explains the exact naming format (dbs-xxx) of the 32 Skills in this project, serving as an auditable naming contract.

2Core Capabilities

01Rule layer audit and classification: determine project category among A/B/C/D and recommend 'which layer to touch first'
02Layered rule file migration: extract platform-agnostic rules into AGENTS.md, leaving Claude-specific ones in CLAUDE.md
03Skill source of truth identification or creation: enter candidate discovery mode when skills/ is absent, producing a candidate inventory tagged with recommended/not recommended for consolidation
04Multi-client bridge generation (Claude / Codex / Grok / general Agents), including Grok's user_invocable hard constraint and precise YAML template
05Naming and frontmatter standardization: kebab-case single name, dbs- prefix, Codex display_name aligned with the standard English name
06Post-migration verification checklist (8 items): AGENTS.md standalone usability, clear source of truth, complete frontmatter, bridges pointing back to source of truth, multi-client set consistency, Grok user_invocable, symlink conflicts, and dangling references

4Recommended Governance & Evidence

Suggested Controls for Safe Operation
For Agent Runtime
Restrict working directory
For Administrators
Retain operation logs
Evidence & Triggers (File Paths & Excerpts)
  • 本地工作目录: File read/write and artifact output: (Instruction-level, non-scripted) Recommends creating/rewriting project AGENTS.md and creating a thin CLAUDE.md when necessary
Audit Notes: . Determination basis: The skill itself contains no scripts, does not access the network, and does not read credentials (the directory only contains SKILL.md and agents/openai.yaml). However, it serves as an operational manual instructing the host agent to modify project structures—involving creating/rewriting AGENTS.md and CLAUDE.md, creating a project-level skills/ directory, generating multi-client bridge files, and creating symlinks in ~/.agents/skills. Thus, its risk surface lies in 'local structure and file read/write operations', with write targets including host skill directories under the user's home directory. SKILL.md employs two confirmation gates to mitigate unintended modifications: every Phase requires 'pre-write confirmation' and stipulates 'only when the user explicitly permits writing to the target host directory may you directly write the bridge to the target location; otherwise, provide a preview first', alongside requiring 'no overwrite' and user confirmation for existing same-name real directories/files before moving them out. No outbound network traffic, no credential access → Classified as Blue tier.

5Second-Pass Verification

  • [ok] 无脚本、无网络、无凭证 — 目录仅 SKILL.md + agents/openai.yaml；token 扫描（curl|wget|http|API_KEY|token|secret|\.env|bash|exec|python|npx）零命中；文中所有「写入」都是给 agent 的指令，不是自带代码。
  • [ok] Grok 硬约束 user_invocable 逐字存在且被反复强调 — grep 命中 8 处，含约束条款、模板内字段、生成期校验、Phase 6 验证项与禁止事项「Grok bridge 绝对不能漏写 user_invocable: true」。
  • [ok] bridge 模板为薄指针、不承载逻辑 — Claude/Codex 模板含 source_of_truth 与 bridge_mode: passthrough，正文仅「请读取真源：…本文件为薄 bridge，仅做入口指向。长期逻辑维护在真源。」；Grok 模板同样只指向 Source of truth。
  • [ok] 「不覆盖既有同名真实目录/文件」是否有落地条款 — 「通用 Agents 目录策略」一节给出三条：不覆盖；报告目标路径和当前类型；让用户确认是否迁出旧目录。Phase 6 另列软链冲突/悬空软链检查项。
  • [ok] 是否承诺执行安装（夸大检查） — SKILL.md 开篇即否认「这不是安装教程。也不是脚本执行器。」；全文无 npx/claude plugin install 之类命令，与自述一致。
  • [ok] 与同仓库既有命名实践一致 — 命名规则第 2 条要求 dbs- 前缀；任务涉及的 8 个 skill 目录名（dbs-action、dbs-ai-check…）与各自 frontmatter name、agents/openai.yaml display_name 三处逐字一致，规则与仓库现状吻合。
  • [ok] 元数据（license/stars/last_push/installs/commit） — GitHub API：stars 10121、pushed_at 2026-09-07T03:48:59Z、license.spdx_id NOASSERTION（根 LICENSE 为 CC BY-NC 4.0，非商业）；本地 HEAD = 8b8e33f1ecaed8cee606fe950c4426b525ead314 = 任务 pin。
  • [discrepancy] meta.installs 值域校正（批4 质检 2026-09-15） — 原记录为 skills.sh 仓库级聚合徽标（579.6K（skills.sh 徽标 /b/dontbesilent2025/…），并非本 skill 的单体安装量；站内惯例只记可定位的单 skill 数值 → 校正为 null（其他 25 份 dbs 报告本就为 null，修此 8 份以齐平）

6Verdict & Scenarios

  • Directly targets real pain points of multi-client Agent workbenches: usable across 4 project types × 5 host states without assuming the project is already standardized.
  • Source of truth/bridge layering provides a sustainably maintainable invariant, preventing identical logic from diverging across four host directories.
  • Host-specific pitfalls are surfaced as generation-time checks (Grok's user_invocable, general Agents' symlink readlink), turning silent failures into verifiable items.
  • Conservative write operations: no overwriting, confirmation required, supports preview-only mode—ideal for operating on projects actively used by others.
  • 6 phases + per-phase confirmation + 5 wrap-up maintenance questions make the delivery process auditable rather than a black-box migration.
    • Best for: Suitable for existing projects of a certain scale with scattered rule files and skills (CLAUDE.md / AGENTS.md / multiple mixed host directories) needing a one-time structural migration; also ideal for developers maintaining skills to be supplied simultaneously to Claude Code, Codex, Grok, and general Agents (Doubao Mac App, Trae Solo)—providing naming contracts, bridge templates, and verification checklists to produce sustainably maintainable workbench structures. Skill authors wanting to learn the engineering practice of 'source of truth + thin bridges for multi-client distribution' will also benefit from reading Phase 5 and the naming rules.
    Not for: Not suitable for scenarios merely wanting to 'copy CLAUDE.md to get Codex up and running' (SKILL.md explicitly states copying does not count as migration and will guide you through a complete round of auditing and naming standardization); not suitable for users expecting a one-click script migration—it deliberately avoids being a script runner and relies entirely on agent-guided phase-by-phase confirmation; not suitable for environments lacking permissions to modify host directories (enterprise-managed devices, shared machines); not suitable for reviewing skill quality or organizing knowledge base content, which falls outside its scope.
    Install Agent-Ready
    ① Verified Mirror 2026-09-15
    Option A · Direct DownloadDownload dbs-agent-migration.tar.gz
    sha256: d64fe9dda1661834…
    Option B · Copy Guide for Agent
    install.json
    Agents automatically download and verify mirrors from this JSON guide.
    ② Upstream GitHub
    GitHub Repo ↗
    Pinned to commit 8b8e33f1ec.