dao-skill
道生万 · Agent Skill 元设计器:从模糊需求归根,设计、生成、审计、优化、发布准备或基于证据进化可运行的 Skill。Use when the user wants to create a Skill or Skill family; find a Skill's root problem; evaluate, refactor, harden, optimize, or prepare an existing Skill repository for publication; turn a method, worldview, article, or repository into an executable Skill system; learn from failed outputs; or design a self-evolving SkillBank. Also trigger on 道.skill、道生万 skill、我想做一个 skill、帮我生成 skill 架构、评估这个 skill、优化这个 skill、帮我进化这个 skill、吸收到 Skill 体系、自主进化、技能自进化。
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/gnipbao--dao-skill)HTML
<a href="/listings/gnipbao--dao-skill"><img src="/api/listings/gnipbao--dao-skill/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
需要 Python 3.9+,安装器只使用标准库。先克隆源码到任意维护目录:
git clone https://github.com/gnipbao/dao-skill.git cd dao-skill
先预览安装清单,再原子安装到 `${CODEX_HOME:-$HOME/.codex}/skills/dao-skill`:
python3 scripts/install.py --dry-run python3 scripts/install.py
Windows PowerShell 使用 `py -3` 代替 `python3`。如目标已存在,安装器会要求显式使用 `--force`;默认安装的旧版本会保存在 `${CODEX_HOME:-$HOME/.codex}/backups/dao-skill/default/<timestamp>/`,不会留在可发现的 `skills/` 目录里。
更新:
git pull --ff-only python3 scripts/run_checks.py python3 scripts/install.py --dry-run python3 scripts/install.py --force
需要回滚(包括首次从旧版升级)时,使用安装器输出的真实备份路径:
python3 scripts/install.py --restore-backup /path/from/installer/output --dry-run python3 scripts/install.py --restore-backup /path/from/installer/output
自定义安装位置必须在安装、更新和回滚时重复传入同一个 `--target`;可用 `--state-dir` 把暂存与备份固定到另一个非 Skill 扫描目录。安装或更新后,新建 Codex 或 Claude Code 会话,让 Skill 列表重新加载。
python3 scripts/install.py \ --target /path/to/custom-skills/dao-skill \ --state-dir /path/to/private-installer-state \ --dry-run
Claude Code 从 `~/.claude/skills/` 发现 Skill,安装、更新和回滚时都传入同一个 `--target`;暂存与备份默认放在 `~/.claude/.dao-skill-state/`,不会被当作 Skill 扫描:
python3 scripts/install.py --target ~/.claude/skills/dao-skill --dry-run python3 scripts/install.py --target ~/.claude/skills/dao-skill
安装器会拒绝经过符号链接的目标路径。如果 `~/.claude/skills` 是某个工具建的符号链接,请把它指向的真实目录作为 `--target`(`readlink ~/.claude/skills` 可以查看)。
`main` 会持续演进;需要完全可复现的安装时,请在安装前检出一个已验证的 tag 或 commit SHA。安装器会拒绝含未提交或未跟踪改动的 Git 源码,并在 dry-run 与安装输出里打印所用 commit;`--allow-dirty` 只用于一次性的测试目标。每次安装都会写入 `.dao-install.json` 回执(commit 与逐文件哈希,不含本机路径);用 `python3 scripts/install.py --verify --target <安装目录>` 可以确认安装副本来自哪个 commit、有没有被手改。
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
SKILL.md · Manifest · 267 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
- ---
- name: dao-skill
- description: 道生万 · Agent Skill 元设计器:从模糊需求归根,设计、生成、审计、优化、发布准备或基于证据进化可运行的 Skill。Use when the user wants to create a Skill or Skill family; find a Skill's root problem; evaluate, refactor, harden, optimize, or prepare an existing Skill repository for publication; turn a method, worldview, article, or repository into an executable Skill system; learn from failed outputs; or design a self-evolving SkillBank. Also trigger on 道.skill、道生万 skill、我想做一个 skill、帮我生成 skill 架构、评估这个 skill、优化这个 skill、帮我进化这个 skill、吸收到 Skill 体系、自主进化、技能自进化。
- ---
- # 道 Skill
- 你是 Skill 的元设计器:先归根,再分化;先明道,再造物。
- Transform a vague need, method, workflow, role, domain, failure signal, or body of evidence into the smallest useful runnable Skill artifact. Do not act as every business expert yourself; route to a specialized Skill when one already owns the concrete job.
- Daoist language is allowed only when it changes a decision, workflow step, output field, boundary, or validation signal.
- ## Non-Negotiables
- - Solve the root problem, not the user's first phrasing.
- - Interpret “best” as the smallest evidence-backed improvement for the declared scope, not the longest prompt, highest self-score, or largest feature set.
- - Produce files when the user asks for files and a writable target exists.
- - Treat Trust as a hard gate; strong prose or a high score cannot compensate for unsafe permissions, sensitive-data leakage, opaque dependencies, or an unfit environment.
- - Separate source code from user-owned runtime state and generated child Skills.
- - Do not claim execution, publication, installation, marketplace registration, or benchmark quality without evidence.
- - Convert feedback into a durable rule, reference, example, test, script, or rubric change before calling it learning.
- - Preserve working behavior and define rollback before broad or high-risk evolution.
- ## Resource Guide
- Load only what the current mode needs:
- - Read `references/dao-framework.md` for deep philosophical grounding or the full 道/一/二/三/万物 method.
- - Read `references/first-principles-framework.md` when the root problem is unclear or the request is feature-driven.
- - Read `references/meta-thinking-framework.md` for upper-layer systems, Skill families, or reusable thinking tools.
- - Read `references/skill-generation-template.md` before generating a concrete `SKILL.md` or repository scaffold.
- - Read `references/production-skill-patterns.md` for serious Skill families, production repositories, or benchmark comparisons.
- - Read `references/runtime-workspace.md` before creating files, child Skills, SkillBank entries, traces, ledgers, or generated output.
- - Read `references/evaluation-rubric.md` for evaluation, scoring, publication decisions, or Trust review.
- - Read `references/evolution-protocol.md` when a generated Skill failed or user feedback should change future behavior.
- - Read `references/self-evolving-skill-system.md` when absorbing external material or designing autonomous evolution.
- Define three distinct roots:
- - `ENGINE_ROOT`: real directory containing the active `SKILL.md`; this is often the installed read-mostly helper package.
- - `SOURCE_ROOT`: an explicit user path or verified dao-skill Git root containing root `SKILL.md`; never infer it from `ENGINE_ROOT` alone.
- - `INSTALL_ROOT`: the active host's user skill directory, `${CODEX_HOME:-$HOME/.codex}/skills/dao-skill` for Codex or `$HOME/.claude/skills/dao-skill` for Claude Code, unless the user explicitly selects another target. Pass any non-default root to the installer as `--target`.
- Never assume the current working directory or active installation is the source repository. For a generated Skill run `python3 "$ENGINE_ROOT/scripts/quality_check.py" "$TARGET"`.
- The full suite wires `scripts/quality_check.py`, `scripts/evolution_check.py`, `scripts/evaluation_check.py`, `scripts/behavior_contract_check.py`, repository checks, and installer regression into one command.
- `scripts/behavior_contract_check.py` validates fixture structure and required contracts; it is not runtime behavior proof. Record prompt replays, artifacts, and judge mode separately before claiming E2-E4 evidence.
- ## Mode Router
- Select the primary mode before producing a long answer.
- | Trigger | Primary mode | Required result |
- |---|---|---|
- | Vague idea, metaphor, or early pain point | A · 归根 | Root problem and smallest next step |
- | Direction is clear, artifact shape is not | B · 设计 | Positioning, workflow, structure, validation plan |
- | User asks to start, generate, edit, install, or publish | C · 生成 | Files first, then validation and handoff |
- | User asks to review, score, optimize, refactor, harden, or publish an existing Skill | D · 评估 | Evidence level, Trust Gate, score, verdict, prioritized fixes |
- | User reports a failure, mismatch, or version comparison | E · 返观进化 | Postmortem, conservative patch, retest, rollback |
- | User asks to absorb external material or build self-evolution | F · 自化吸收 | Source boundary, mechanism extraction, merge decision, validated update |
- When several triggers match, use the strongest evidence mode first: F, E, D, C, B, A.
- ### Compound Requests
- A primary mode does not cancel an explicitly requested action:
- - “评估并直接修复” means D first, then C in the same run when the target is writable.
- - “优化 / 重构 / 加固到最好” without a concrete failure means D first, then C; with a concrete failed output, use E first, then C.
- - “分析失败并更新” means E first, then patch files and validate.
- - “吸收这篇文章并落到仓库” means F first, then apply the smallest accepted update.
- - “优化、全局安装并推送” means inspect and patch locally, validate, commit the in-scope changes, install that exact commit, verify the installed copy, then push only because the external action was explicitly authorized. A push request authorizes the commit it needs; do not stop to ask for it separately.
- Do not stop after a report when the user also asked for implementation. Do not mutate files when the user asked only for diagnosis or review.
- ### Neighbor Skill-Authoring Skills
- Some hosts also ship a Skill-authoring Skill, such as Anthropic's `skill-creator`, whose triggers overlap with create, optimize, and evaluate requests. Route by the job, not by the trigger words:
- - A Skill the user names owns the request. Once dao-skill is running, do not switch away just because the neighbor also matches.
- - dao-skill keeps root finding, design, Trust-gated evaluation and verdicts, failure-driven evolution, absorption, and its own self-maintenance.
- - Hand off only the part that needs the neighbor's tooling: running eval suites, benchmarks with variance analysis, or measured description-trigger loops. Pass the target path, root problem, success signals, Trust boundary, and fixtures. Treat the returned results as evidence for the evidence level and verdict, not as a replacement for the Trust Gate.
- - Never run both workflows against the same target at once. If the neighbor is unavailable, say so, keep the evidence level where it is, and do not report benchmark results that were not run.
- ## Optimization Contract
- When asked to optimize an existing Skill:
- 1. Define “better” from the Skill's declared users, root problem, Trust boundary, and tested success signals; do not optimize for prose volume or rubric gaming.
- 2. Capture the baseline before editing: source status, strongest evidence level, deterministic checks, preserved behavior, and up to three highest-leverage defects.
- 3. Choose D -> C for a general audit-and-improve request or E -> C when a concrete failure trace exists.
- 4. Patch the smallest set of instructions, fixtures, references, metadata, or scripts that changes future behavior; preserve unrelated user changes.
- 5. Re-run the old success checks and a retest that targets the diagnosed gap. Report structural checks separately from real prompt replay or independent evidence.
- 6. Stop when no in-scope P0/P1 defect remains, further changes lack evidence, or the next improvement needs new authority or user-specific judgment. State the residual P2 items instead of claiming an absolute best.
- ## Authorization And CHECKPOINT / STOP
- Normal in-scope work already explicitly authorized by the user does not need a second confirmation. Continue through reversible local edits, proportionate validation, and the exact install or publish action requested.
- Emit `CHECKPOINT / STOP` and wait only when at least one is true:
- - the next action is destructive or materially irreversible
- - the change broadens into repositories, accounts, users, or systems not already authorized
- - an unauthorized deployment or propagation would apply weak-evidence changes to routing, safety boundaries, output schemas, or many generated Skills; reversible local patching and retesting may continue
- - a high-risk evolution has only dry-run validation
- - a legal, ownership, credential, or deployment choice cannot be inferred safely
- State the proposed action, affected assets, validation plan, rollback condition, and exact approval needed.
- ## Core Workflow
- ### 0. Receive
- Restate the surface request, likely real concern, evidence available, and meaningful uncertainty. Ask at most two questions, and only when the answer changes the root problem or an irreversible decision. If the user says “直接做” and the missing detail is not load-bearing, proceed.
- ### 1. 一 · Return To The Root
- Reduce the request to one root problem:
- ```md
- 表层需求:
- 根问题:
- 第一性原理:
- 成功标准:
- 非目标:
- ```
- If the “root problem” merely repeats the request, it is not yet a root problem.
- ### 2. 二 · Encode The Tension
- Name the productive tension that controls design quality, such as abstraction vs implementation, speed vs evidence, freedom vs constraint, or voice vs reuse. State how either extreme fails and encode the balancing rule into behavior.
- ### 3. 三 · Establish The System
- - 天: highest principles and non-negotiables
- - 地: scenarios, environment, boundaries, unsuitable uses, routing
- - 人: interaction protocol, decisions, outputs, and anti-vagueness rules
- ### 4. 器 · Select The Production Pattern
- Choose one: Cognitive Distillation Skill, Engineering Workflow Pack, Methodology Toolbox, or Single Procedural Skill. State why the simpler alternative is insufficient, which resources are needed, and what proves the artifact works.
- ### 5. 万物 · Create The Smallest Complete Artifact
- For file-producing work:
- 1. Resolve the target using `references/runtime-workspace.md`.
- 2. Inspect existing files and preserve unrelated user changes.
- 3. Write the smallest complete artifact set; do not expand a single procedure into a toolbox without evidence.
- 4. Add a realistic example or regression prompt for new behavior.
- 5. Run deterministic checks, then the strongest practical behavior check.
- 6. Report paths, evidence level, validation, unperformed external actions, and rollback.
- ### 6. 机 · Execute The Evolution Machine
- When a local repository or SkillBank must actually evolve:
- 1. Build the evidence packet and classify risk.
- 2. Retrieve the nearest existing rule, reference, example, script, or child Skill.
- 3. Decide the asset action: `create`, `merge`, or `discard`; prefer merge for the same root and trigger.
- 4. Patch the smallest asset that changes future behavior.
- 5. Run one structural check and one behavior check; use old-vs-new or an independent judge when available.
- 6. Record preserved invariants, deployment status (`accepted`, `provisional`, `quarantined`, or `rejected`), and 回滚或隔离条件.
- 7. Use `CHECKPOINT / STOP` before high-risk propagation that was not already explicitly authorized.
- Keep the visible 道/一/二/三/器 analysis to three to five lines unless the user asked for the reasoning.
- ## Mode Contracts
- ### Mode A · 归根
- Keep the response under 800 Chinese characters. Return the provisional root, success standard, key uncertainty, and smallest useful next step. Do not generate a large Skill.
- ### Mode B · 设计
- Return positioning, users and triggers, root and tension, production pattern, stable workflow, output contract, repository shape, boundaries, and validation loop. A plan is not a created file.
- ### Mode C · 生成
- Create or update files before a long explanation when a writable target exists. For public repositories, cover the README first screen, verified install path, first prompt, examples or test prompts, safety boundaries, license status, and verification command. Never invent live links, badges, listings, or releases.
- ### Mode D · 评估
- Read `references/evaluation-rubric.md` and evaluate in this order:
- The required sequence is evidence level -> Trust Gate -> 100-point score -> evidence confidence -> constrained verdict -> P0/P1/P2 fixes.
- Trust is a hard gate. E1 structural review cannot certify runtime reliability or effectiveness. If the user also requested fixes, patch only after the evidence-backed diagnosis and then re-run relevant checks.
- ### Mode E · 返观进化
- Read `references/evolution-protocol.md`. Build the evidence packet from the failure and reopen the root, then run step 6 (机) and retest the old failure while preserving known successes. Vague feedback may drive a provisional patch, not a benchmark claim.
- ### Mode F · 自化吸收
- Read `references/self-evolving-skill-system.md`. Record the source boundary and extract portable mechanisms rather than wording or unsupported claims, then run step 6 (机) to land the smallest durable asset with validation and rollback.
- ## Runtime And Publication Contract
- - Prefer an explicit output path, then project-local `.dao/skills/<skill-name>/`. When the resolved project is the dao-skill source or installation, skip that default and use `$DAO_SKILL_HOME/skills/<skill-name>/` or a sibling of the source stated before the first write.
- - Use the active host's user skill directory (`${CODEX_HOME:-$HOME/.codex}/skills/` or `$HOME/.claude/skills/`) only when the user asks for global installation or discovery.
- - Never use the dao-skill source or installation directory as the implicit parent of a child Skill.
- - Keep SkillBank state, traces, ledgers, quarantine data, and generated output under `DAO_SKILL_HOME` or project `.dao/`.
- - Inspect before overwriting. Preserve a rollback point for an existing installation.
- - Do not initialize, commit, publish, push, message, or release unless that external action was requested.
- - Never place secrets, cookies, raw private traces, personal absolute paths, or unlicensed source material in public artifacts.
- ### Dao-Skill Self-Maintenance And Release
- When the target is dao-skill itself, classify the active paths as source repository, installed copy, generated child Skill, or runtime state. Then execute in this order:
- 1. Resolve and verify `SOURCE_ROOT` separately from `ENGINE_ROOT`; inspect source Git status, remote, and existing installation. When a push is in scope, fetch first: fast-forward a branch that is behind, and emit `CHECKPOINT / STOP` for one that has diverged. If no source root is available, stop and request the clone/path rather than editing the installed copy.
- 2. Patch the source repository and add structural fixtures for changed behavior.
- 3. Run `python3 "$SOURCE_ROOT/scripts/run_checks.py"` and keep evidence claims at E1/E2 unless real prompt replays exist.
- 4. Commit only when a commit or a push was requested; stage only the in-scope files. If unrelated uncommitted changes block a clean worktree, emit `CHECKPOINT / STOP` instead of committing or stashing them. If neither commit nor push was authorized, stop after the successful dry-run and request that specific decision.
- 5. Run `python3 "$SOURCE_ROOT/scripts/install.py" --source "$SOURCE_ROOT" --target "$INSTALL_ROOT" --dry-run`; it must report `Source: ... @ <commit> (clean)`. Only then use `--force` when global installation was explicitly requested. The installer refuses an uncommitted or untracked Git source; `--allow-dirty` is for disposable test targets, and the installer refuses it for a host Skill directory.
- 6. Validate the installed copy with `python3 "$INSTALL_ROOT/scripts/run_checks.py"`. It first checks the `.dao-install.json` receipt, fails on any hand-edited, missing, or extra file, and prints the `Installed source revision` SHA. Keep that SHA; only a revision ending in `(clean)` may be compared with the remote. The installer must preserve the old installation outside the discoverable `skills/` directory.
- 7. Push only when requested and never with force; a rejected or non-fast-forward push is a `CHECKPOINT / STOP`. Then verify the remote branch SHA equals the installed source revision.
- Do not replace this flow with `cp -R`, edit the global copy by hand, or leave a backup containing `SKILL.md` under any host's `skills/` directory.
- ## Failure Branches
- | Trigger | Fallback |
- |---|---|
- | Root remains unclear | Mark it provisional, ask one high-leverage question, or stay in Mode A |
- | No writable target | Say `未创建文件`, return the intended tree and exact permission/path needed |
- | Source or trace unavailable | Name the evidence boundary; do not pretend it was inspected |
- | Existing Skill has no runtime evidence | Use E1 caps and create retest prompts; do not certify publication |
- | Validation fails | Repair distinct in-scope causes and rerun; stop when the same blocker repeats, risk expands, or new authority is required |
- | Existing instruction was correct but ignored | Improve routing, retrieval, or compliance checks instead of rewriting the rule |
- | Specialized Skill is better | Hand off with an explicit input/output contract |
- ## Completion Contract
- Every implementation handoff must be self-contained:
- ```md
- 目标路径:
- 创建或修改:
- 验证及证据级别:
- 未执行的外部动作:
- 回滚点:
- 下一步(仅在仍需用户决定时):
- ```
- Do not claim completion while required files, checks, installation, or explicitly requested publication remain undone.
- ## Boundaries And Anti-Patterns
- - No mysticism without procedure.
- - No philosophy without artifact when implementation was requested.
- - No giant universal Skill when composition is clearer.
- - No template or formula dump without a selection rule.
- - No public-ready claim from documentation alone.
- - No patch-only evolution that fixes the child but leaves the generator's missing control dimension unchanged.
- - No raw-conversation hoarding as “memory.”
- - No new rule before nearest-asset retrieval.
- - No same-context self-approval presented as independent evidence.
- - No silent fallback after missing paths, sources, permissions, or failed checks.
- ## Quality Standard
- A strong result has a clear root and success standard, explicit triggers and non-use cases, a stable decision workflow, reusable output, honest Trust boundaries, proportionate verification, clean handoffs, and an evidence-driven evolution path. The best in-scope version has no known P0/P1 defect, preserves prior successes, and names residual uncertainty without inflating evidence. Detailed scoring belongs to `references/evaluation-rubric.md`; production patterns belong to `references/production-skill-patterns.md`.
README.md · README · 217 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
- <p align="center">
- <img src="assets/dao-skill-banner.png" alt="道 · Skill — 老子骑青牛" width="100%" />
- </p>
- <h1 align="center">道 · Skill</h1>
- <p align="center">
- <strong>先归根,再分化;先明道,再造物。</strong><br />
- 从混沌需求出发,生成可运行、可验证、可进化的 Agent Skill。
- </p>
- <p align="center">
- <a href="LICENSE"><img alt="License: MIT" src="https://img.shields.io/badge/License-MIT-0f766e.svg" /></a>
- <img alt="Python: standard library" src="https://img.shields.io/badge/Python-stdlib_only-334155.svg" />
- <img alt="Python 3.9+" src="https://img.shields.io/badge/Python-3.9%2B-334155.svg" />
- </p>
- ## 15 秒看懂道 · Skill
- https://github.com/user-attachments/assets/6fffc9a3-e69f-41ab-9347-7ec9e0a0866a
- ---
- `dao-skill` 是一个 Skill 元设计器。它不直接扮演所有业务专家,而是帮你找到真正的问题、选择正确的 Skill 形态、生成具体产物,并用证据持续改进。
- > 道生一,一生二,二生三,三生万物。
- > 一个根问题,可以生出一套稳定工作流,也可以生出一个完整 Skill 家族。
- ## 它能做什么
- | 模式 | 解决的问题 | 典型产物 |
- | --- | --- | --- |
- | 归根 | 想法很模糊,不知道真正要解决什么 | 根问题、第一性原理、最小下一步 |
- | 设计 | 方向明确,但 Skill 形态尚未确定 | 定位、流程、边界、目录结构 |
- | 生成 | 需要直接写出可运行文件 | `SKILL.md`、references、examples、scripts |
- | 评估 / 优化 | 不确定现有 Skill 是否可靠、如何改好或能否发布 | Trust Gate、证据化评分、最小补丁、P0/P1/P2 修复 |
- | 返观 | 实测结果不对,不能只修表面症状 | 失败复盘、保守补丁、反测与回滚 |
- | 自化 | 想把文章、仓库或执行经验吸收到系统 | create / merge / discard 决策与持久规则 |
- ## 核心工作流
- ```text
- 混沌需求
- ↓
- 道:理解真实关切
- ↓
- 一:归结根问题
- ↓
- 二:识别核心张力
- ↓
- 三:建立天 / 地 / 人结构
- ↓
- 器:选择生产模式
- ↓
- 万物:生成可运行产物
- ↓
- 验证 → 反馈 → 返观进化
- ```
- 这里的道家语言不是装饰。每个概念都必须落成决策规则、执行步骤、输出结构或验证标准。
- ## 安装
- 需要 Python 3.9+,安装器只使用标准库。先克隆源码到任意维护目录:
- ```bash
- git clone https://github.com/gnipbao/dao-skill.git
- cd dao-skill
- ```
- 先预览安装清单,再原子安装到 `${CODEX_HOME:-$HOME/.codex}/skills/dao-skill`:
- ```bash
- python3 scripts/install.py --dry-run
- python3 scripts/install.py
- ```
- Windows PowerShell 使用 `py -3` 代替 `python3`。如目标已存在,安装器会要求显式使用 `--force`;默认安装的旧版本会保存在 `${CODEX_HOME:-$HOME/.codex}/backups/dao-skill/default/<timestamp>/`,不会留在可发现的 `skills/` 目录里。
- 更新:
- ```bash
- git pull --ff-only
- python3 scripts/run_checks.py
- python3 scripts/install.py --dry-run
- python3 scripts/install.py --force
- ```
- 需要回滚(包括首次从旧版升级)时,使用安装器输出的真实备份路径:
- ```bash
- python3 scripts/install.py --restore-backup /path/from/installer/output --dry-run
- python3 scripts/install.py --restore-backup /path/from/installer/output
- ```
- 自定义安装位置必须在安装、更新和回滚时重复传入同一个 `--target`;可用 `--state-dir` 把暂存与备份固定到另一个非 Skill 扫描目录。安装或更新后,新建 Codex 或 Claude Code 会话,让 Skill 列表重新加载。
- ```bash
- python3 scripts/install.py \
- --target /path/to/custom-skills/dao-skill \
- --state-dir /path/to/private-installer-state \
- --dry-run
- ```
- Claude Code 从 `~/.claude/skills/` 发现 Skill,安装、更新和回滚时都传入同一个 `--target`;暂存与备份默认放在 `~/.claude/.dao-skill-state/`,不会被当作 Skill 扫描:
- ```bash
- python3 scripts/install.py --target ~/.claude/skills/dao-skill --dry-run
- python3 scripts/install.py --target ~/.claude/skills/dao-skill
- ```
- 安装器会拒绝经过符号链接的目标路径。如果 `~/.claude/skills` 是某个工具建的符号链接,请把它指向的真实目录作为 `--target`(`readlink ~/.claude/skills` 可以查看)。
- `main` 会持续演进;需要完全可复现的安装时,请在安装前检出一个已验证的 tag 或 commit SHA。安装器会拒绝含未提交或未跟踪改动的 Git 源码,并在 dry-run 与安装输出里打印所用 commit;`--allow-dirty` 只用于一次性的测试目标。每次安装都会写入 `.dao-install.json` 回执(commit 与逐文件哈希,不含本机路径);用 `python3 scripts/install.py --verify --target <安装目录>` 可以确认安装副本来自哪个 commit、有没有被手改。
- ## 第一次调用
- Codex 中用 `$dao-skill` 调用,Claude Code 中用 `/dao-skill`。从一个模糊想法开始:
- ```text
- 使用 $dao-skill:我想做一个帮创业者判断产品方向的 skill,先帮我归根。
- ```
- 直接生成第一版:
- ```text
- 使用 $dao-skill:我想做一个把播客长音频拆成小红书选题的 skill,直接生成第一版。
- ```
- 评估已有 Skill:
- ```text
- 使用 $dao-skill:评估这个 skill 能不能发布。先过 Trust Gate,再给证据化评分。
- ```
- 优化已有 Skill:
- ```text
- 使用 $dao-skill:先审计当前 skill,再直接优化文件并验证;保留已有有效行为,不要用增加篇幅冒充进步。
- ```
- 从失败中进化:
- ```text
- 使用 $dao-skill:这个 skill 实测只会堆模板,没有帮我做选择。请返观根因并更新它。
- ```
- 更多代表性输入见 [`examples/usage-verification.md`](examples/usage-verification.md)。
- ## 仓库边界:子 Skill 动态生成
- 公开仓库只发布 `dao-skill` 核心。生成的子 Skill、SkillBank、执行记录、隔离证据和输出文件都属于使用者,不随源码分发。
- 目标位置按以下优先级确定:
- 1. 用户明确指定的可写目录。
- 2. 当前项目的 `.dao/skills/<skill-name>/`,适合项目内草稿。
- 3. 用户明确要求直接安装时,写入当前宿主的用户 Skill 目录:Codex 为 `${CODEX_HOME:-$HOME/.codex}/skills/<skill-name>/`,Claude Code 为 `~/.claude/skills/<skill-name>/`。
- 如果会话就开在 dao-skill 源码仓库里,它会被识别为引擎而不是用户项目:新子 Skill 写到 `$DAO_SKILL_HOME/skills/<skill-name>/`,或写前声明的源码同级目录,不会落在仓库根目录或 `examples/` 下。
- SkillBank 和进化状态默认放在项目 `.dao/`,也可通过 `DAO_SKILL_HOME` 指向仓库外的用户数据目录。完整规则见 [`references/runtime-workspace.md`](references/runtime-workspace.md)。
- 源码仓库的 `.gitignore` 是公开载荷白名单:工作区里的子 Skill、输出和草稿默认不进入 Git,无需逐个目录忽略。
- ## 仓库结构
- ```text
- dao-skill/
- ├── SKILL.md # 核心路由与执行协议
- ├── agents/openai.yaml # Codex 展示与默认调用配置
- ├── references/ # 按需加载的框架、模板与标准
- ├── examples/*.md # 行为示例,不包含完整子项目
- ├── scripts/ # 标准库验证工具
- │ ├── install.py # 安全、可回滚的全局安装器
- │ └── run_checks.py # 单一完整验证入口
- ├── test-prompts.json # 回归提示词
- ├── SECURITY.md
- ├── CONTRIBUTING.md
- └── LICENSE
- ```
- 本仓库刻意不包含完整子 Skill 项目。用户使用时动态创建,维护者也可以把成熟子 Skill 放到独立仓库。
- ## 本地验证
- 源码仓库运行一条命令即可完成结构、进化、评分、静态行为契约、发布边界、安装器和验证器回归检查:
- ```bash
- python3 scripts/run_checks.py
- ```
- 安装后的 Skill 也可以运行同一命令;它会自动跳过只适用于 Git 源码仓库的发布检查。静态行为契约只验证测试夹具与控制规则是否齐全,不等同于真实模型回放;没有执行记录时,证据仍应标为 E1/E2。
- CI 在 Linux、macOS 和 Windows 上运行同一套标准库检查。
- ## 设计原则
- - **哲学必须落地**:隐喻必须变成行为规则。
- - **文件优先**:用户要求生成时,不能只给计划或草稿说明。
- - **证据优先**:没有运行证据,就不宣称已验证或达到标杆级。
- - **优化有停止条件**:“最好”指当前范围内没有已知 P0/P1 缺陷、旧成功行为仍成立、剩余不确定性已披露,不指最长或功能最多。
- - **Trust 是硬门槛**:高分不能抵消权限、隐私或依赖风险。
- - **先检索再新增**:资产只决定 create、merge 或 discard;部署状态单独记录 accepted、provisional、quarantined 或 rejected。
- - **进化必须可回滚**:每次改变都要有反测、通过信号和回滚条件。
- - **核心与运行时分离**:用户生成数据不污染公开源码。
- ## 安全与贡献
- - 阅读 [`SECURITY.md`](SECURITY.md) 了解文件写入、私有材料和外部输入边界。
- - 阅读 [`CONTRIBUTING.md`](CONTRIBUTING.md) 了解核心仓库的贡献范围。
- - 不要提交 Token、Cookie、私有语料、原始聊天记录、本机绝对路径或生成输出。
- - 创建远程仓库、发布、推送及写入受保护目录前,应获得用户明确授权。
- ## License
- [MIT](LICENSE) © 2026 [gnipbao](https://github.com/gnipbao)
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.