Skip to content
Back to search

first-look

SkillWorks with Claude CodeGitHub

Pre-release repository audit and cold-start verification from a stranger's first-visit perspective (仓库发布前外部视角体检与冷启动验收). Default Light mode (read-only, no installs or large downloads): one-sentence verdict, staged delivery boundaries (bundled vs user-supplied vs failure mode), external exposure audit across working tree, publishable Git history, archives, metadata, and third-party licenses, plus safe public-claims guardrails. Explicit Deep mode (only when user asks to run/install): isolated cold-start execution on the actual release artifact. Not for unit tests, code review, pentesting, or legal sign-off.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/seanyuanwsy--first-look/badge.svg)](/listings/seanyuanwsy--first-look)

HTML

<a href="/listings/seanyuanwsy--first-look"><img src="/api/listings/seanyuanwsy--first-look/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Follows the standard [Agent Skills](https://agentskills.io/) layout (`SKILL.md` + `references/`). Compatible with Claude Code, Codex, Cursor, OpenCode, Oh My Pi, and other Agent Skills hosts.

Option 1 — Skills CLI

npx skills add SeanYuanWSY/first-look

Option 2 — Git clone (global or project-local)

# Universal shared skills directory (Codex, Oh My Pi, etc.)
git clone https://github.com/SeanYuanWSY/first-look.git ~/.agents/skills/first-look

# Or Claude Code personal skills directory
git clone https://github.com/SeanYuanWSY/first-look.git ~/.claude/skills/first-look

Uninstall

rm -rf ~/.agents/skills/first-look   # or ~/.claude/skills/first-look

> **Platform note**: Shell snippets in [`references/checklist.md`](references/checklist.md) target POSIX shells (macOS / Linux); Windows users can use the noted PowerShell equivalents or tool-neutral steps.

---

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 · 62 lines
  1. ---
  2. name: first-look
  3. description: >-
  4. Pre-release repository audit and cold-start verification from a stranger's first-visit
  5. perspective (仓库发布前外部视角体检与冷启动验收). Default Light mode (read-only, no installs
  6. or large downloads): one-sentence verdict, staged delivery boundaries (bundled vs
  7. user-supplied vs failure mode), external exposure audit across working tree, publishable
  8. Git history, archives, metadata, and third-party licenses, plus safe public-claims
  9. guardrails. Explicit Deep mode (only when user asks to run/install): isolated cold-start
  10. execution on the actual release artifact. Not for unit tests, code review, pentesting, or
  11. legal sign-off.
  12. license: MIT
  13. metadata:
  14. version: "1.0.0"
  15. ---
  16. # First Look:外部视角发布前体检与冷启动验收
  17. 以**第一次点进仓库的外部用户**视角回答三件事:**外人看它是什么、公开交付带了什么与缺了什么、哪些未计划公开的内容已经暴露**。
  18. 默认**只查不改、不自动清理、不自动发布**;不替代单元测试、代码审查、安全渗透或法律合规评审。
  19. **语言规则**:使用用户的提问语言回复;英文环境下使用文末中英对照标签,不混用语言。
  20. ## 1. 两档模式(默认轻档)
  21. - **轻档(默认:交付边界 + 暴露面体检)**:只读检查拟发布工件、首屏文档、关键入口与将公开的 Git 历史/元数据;**不安装依赖、不下载大包、不展开内部函数实现或运行日志**。开工前**必须读取** [references/checklist.md](references/checklist.md) 执行只读排查;首次出报告前按需对照 [references/examples.md](references/examples.md)。
  22. - **深档(冷启动实跑验收)**:**仅当用户明确要求实跑/安装验证时启用**。必须在一次性隔离环境中对**实际拟交付渠道工件**(保留会随件发出的文件)执行安装与 1–2 个文档宣称的核心任务;若无法隔离或仓库缺少安装文档,直接按第 5 节停步,绝不在宿主环境硬跑或凭记忆补救。
  23. ## 2. 观察者纪律、范围锁与证据等级
  24. 1. **一行锁定对象**:首句用固定枚举标注 `对象@版本/提交〔渠道|候选或线上|静态或实跑〕`(渠道如 `Git 树` / `源码包` / `Wheel` / `Release 附件` / `容器镜像`);大型 monorepo 须写明受审子路径,未查路径与未生成产物标 `未核实`。
  25. 2. **证据隔离**:作者本人也能跑轻档。先只凭拟发布工件描述陌生人看到的事实,再按用户明示的发布意图判偏离。**将公开的 Git 历史可作曝光证据**;旧会话、私人交接、兄弟仓库不得用来补全外部用户认知或运行条件,也不写删掉零损失的旧上下文否定句(“它不是 X”)。
  26. 3. **证据等级严格分开**:只用 `文档声明` / `静态核实` / `实跑核实` / `未核实`。轻档绝不宣称“实际跑通”,只写“公开交付提供到哪一步、首个前置条件是什么”。
  27. 4. **源头脱敏**:扫描命令在进入终端或工具日志前就必须脱敏,**严禁把疑似密钥、令牌或明文私人邮箱原值打进日志或报告**,只报位置/提交与信息类型。
  28. ## 3. 轻档检查要点(执行细则见 `references/checklist.md`)
  29. - **首屏与定位**:核对 README 首屏、快速开始可见性、许可证文件与仓库元信息。
  30. - **真实交付边界(逐阶段拆 5 层)**:分开核对 `代码` / `解释器与环境` / `权重与数据` / `许可与平台权限` / `网络与凭据`;缺失后果标 `硬停 / 降级 / 替代 / 未知`,标清证据等级与出处。
  31. - **外部暴露面(三层可见性分开标)**:严格区分 `仅本地` / `拟随发布可见` / `线上已可见`。覆盖:
  32. - **可定位的 AI/Agent 协作痕迹**(如 `.claude/`、`CLAUDE.md`、`AGENTS.md`、`GEMINI.md`、`.cursor/`、`.cursorrules`、`.github/copilot-instructions.md`、`.aider*`、`.codex/`、`.kiro/`、`.ai/` 或会话注释):**只报可定位文件/行号,绝不凭“猜文风”列发现**;如实报为可见事实,**不把主动公开的 Agent 配置默认当缺陷**,仅对照作者发布意图判偏离。
  33. - **内部计划、私有与隐藏信息**:未发布路线图/内部编号、拟推送历史中的 Author/Committer 邮箱与已删文件、本机路径、疑似密钥、锁文件/子模块/LFS/归档/容器层/notebook/SVG 元数据、私人 fork 链接。
  34. - **多许可证与随包第三方资产**:排查根及嵌套 `LICENSE`/`COPYING`/`NOTICE`、SPDX 字段及随包第三方代码/权重/数据/图件,分开判断 `可下载 / 可使用 / 可随包再分发`;冲突或不清标 `未知`。
  35. ## 4. 轻档固定 4 块输出(短句直给,高危前置)
  36. 1. **一句判断**:`对象@版本〔渠道|候选/线上|静态〕` 外人看它是 ___,公开交付提供到 ___,首个前置条件是 ___(如有发布阻断级硬停或高危泄露,此句末尾直接点明)。
  37. 2. **实际交付表**:`阶段 | 随件已带 | 另需自备(标层级) | 缺失后果 | 依据`(只列主路径与影响结论的关键分支)。
  38. 3. **外部暴露面表(按风险从高到低排)**:`风险 | 发现 | 外人会知道什么(直接事实;推断另标) | 位置/渠道 | 可见状态 | 处置建议`。表后一行写明已查范围、扫描限制与未覆盖项。
  39. 4. **对外口径红线**:各列 1–3 条极短句的 `对外可写` 与 `证据不足,暂勿写`。
  40. > 除上述 4 块外不另开长篇章节;允许在表 3 的 `处置建议` 列给出一句话建议,但不自动修改文件。
  41. ## 5. 深档补充规则(仅用户明确要求实跑时执行)
  42. - **强制安全隔离**:在一次性容器/VM(空 `HOME`、无宿主凭据/SSH/容器套接字、受限出站与资源)中执行;解包归档前先校验路径穿越与越界符号链接。**宿主无法提供有效隔离时,直接停止并返回 `证据不足`,严禁在宿主硬跑不受信任安装钩子。**
  43. - **零文档停步**:仓库缺少可读 README 或安装说明时,将“缺少安装文档”记为首个硬停并跳过实跑,不凭经验猜测补齐。
  44. - **证据绑定与脱敏**:每条 `实跑核实` 必须绑定到实际安装渠道、不可变摘要/提交、执行任务与公开前置条件,不得把源码检出实跑泛化到 Wheel/容器等其他渠道;截图与日志附证前须脱敏;结束后清理临时环境并报告清理结果。
  45. - **深档输出**:在轻档 4 块后仅追加 `实跑结果与真实阻塞` 一节,给出冷启动结论:`通过 / 部分通过 / 不通过 / 证据不足`。
  46. ## 6. 中英标签对照(按用户语言择一)
  47. `候选 / 线上` ↔ `Candidate / Live` · `静态 / 实跑` ↔ `Static / Executed` · `硬停 / 降级 / 替代 / 未知` ↔ `Hard stop / Fallback / Alternative / Unknown` · `文档声明 / 静态核实 / 实跑核实 / 未核实` ↔ `Doc claim / Static verified / Executed / Unverified` · `仅本地 / 拟随发布可见 / 线上已可见` ↔ `Local only / Planned public / Already public` · `高 / 中 / 低` ↔ `High / Medium / Low` · `通过 / 部分通过 / 不通过 / 证据不足` ↔ `Pass / Partial / Fail / Insufficient evidence`
README.md · README · 96 lines
  1. <p align="center">
  2. <img src="assets/hero.png" alt="first-look — Pre-release Repo Audit for AI Agents" width="100%" />
  3. </p>
  4. <h1 align="center">first-look</h1>
  5. <p align="center">
  6. <strong>A pre-release repository audit &amp; cold-start check skill for AI coding agents.</strong><br />
  7. See your repository through a first-time visitor's eyes: what it appears to be, what it actually ships vs. misses, and what unintended traces are already exposed.
  8. </p>
  9. <p align="center">
  10. <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-0EA5E9.svg?style=flat-square" alt="MIT License" /></a>
  11. <a href="https://agentskills.io/"><img src="https://img.shields.io/badge/Agent_Skills-compatible-10B981.svg?style=flat-square" alt="Agent Skills Compatible" /></a>
  12. <a href="CHANGELOG.md"><img src="https://img.shields.io/badge/version-1.0.0-38BDF8.svg?style=flat-square" alt="Version 1.0.0" /></a>
  13. </p>
  14. <p align="center">
  15. <strong>English</strong> | <a href="README.zh-CN.md">简体中文</a>
  16. </p>
  17. When you know a repository inside out, it is surprisingly easy to ship hidden assumptions ("one command runs everything" when a model weight or separate runtime environment is missing) or unintended traces (personal commit emails, internal project codenames, draft roadmaps, or leftover agent files).
  18. **`first-look`** gives any Agent Skills–compatible coding agent a disciplined, read-only **external-visitor lens** before you publish a repository or whenever you inspect an unfamiliar one.
  19. ---
  20. ## Two Modes
  21. | Mode | Trigger | What it does |
  22. | --- | --- | --- |
  23. | **Light (default)** | Any normal first-look / pre-release check | **Read-only.** Never installs packages or downloads large files. Outputs 4 compact blocks: **1-sentence verdict**, **5-layer delivery boundary table**, **risk-ranked external exposure table**, and **public claims guardrail**. |
  24. | **Deep (explicit)** | Only when you explicitly ask to install & run | Runs the actual release artifact inside a **disposable sandbox** (empty `HOME`, stripped credentials, verified archive paths) and binds every `Executed` verdict to the exact channel and digest. Refuses to execute on the host if isolation is unavailable. |
  25. ### What Light Mode Audits
  26. - **True delivery boundaries (5 layers per stage)**: Code, Interpreter/Runtime, Weights/Data, Licenses/Platform permissions, and Network/Credentials — with explicit failure modes (`Hard stop` / `Fallback` / `Alternative` / `Unknown`) and evidence levels (`Doc claim` / `Static verified` / `Executed` / `Unverified`).
  27. - **External exposure (3 visibility tiers)**: Separates `Local only`, `Planned public` (tracked tree, full reachable Git history, archives/packages), and `Already public` (remote/CI/releases/forks).
  28. - Source-level redaction: scanners never dump raw secrets, personal emails, or local usernames into tool logs.
  29. - Evidence-only AI/agent detection: reports concrete files/lines neutrally (never guesses authorship from prose style, and never flags intentionally published `AGENTS.md`/`CLAUDE.md` as defects by default).
  30. - Multi-license & third-party redistribution checks: separates **downloadable**, **usable**, and **redistributable**.
  31. ---
  32. ## Installation
  33. Follows the standard [Agent Skills](https://agentskills.io/) layout (`SKILL.md` + `references/`). Compatible with Claude Code, Codex, Cursor, OpenCode, Oh My Pi, and other Agent Skills hosts.
  34. ### Option 1 — Skills CLI
  35. ```sh
  36. npx skills add SeanYuanWSY/first-look
  37. ```
  38. ### Option 2 — Git clone (global or project-local)
  39. ```sh
  40. # Universal shared skills directory (Codex, Oh My Pi, etc.)
  41. git clone https://github.com/SeanYuanWSY/first-look.git ~/.agents/skills/first-look
  42. # Or Claude Code personal skills directory
  43. git clone https://github.com/SeanYuanWSY/first-look.git ~/.claude/skills/first-look
  44. ```
  45. ### Uninstall
  46. ```sh
  47. rm -rf ~/.agents/skills/first-look # or ~/.claude/skills/first-look
  48. ```
  49. > **Platform note**: Shell snippets in [`references/checklist.md`](references/checklist.md) target POSIX shells (macOS / Linux); Windows users can use the noted PowerShell equivalents or tool-neutral steps.
  50. ---
  51. ## Usage Examples
  52. - *"Run a `first-look` check on this repo before I make it public."*
  53. - *"Use `first-look` to audit this repository for missing runtime prerequisites or exposed internal traces."*
  54. - *"Run `first-look` in **deep mode** on `dist/mypkg-1.0.0.tar.gz` in a clean container."*
  55. The agent replies in the language of your prompt using consistent English or Chinese status labels.
  56. ---
  57. ## Scope & Non-Goals
  58. - **Read-only by default**: inspects and recommends; never edits files, rewrites Git history, or publishes on its own.
  59. - **Not a replacement** for unit tests, code review, penetration testing, or legal compliance sign-off.
  60. ---
  61. ## Repository Layout
  62. - [`SKILL.md`](SKILL.md) — Core skill specification (modes, observer isolation, 4-block schema, deep-mode safety contract, CN/EN label map)
  63. - [`references/checklist.md`](references/checklist.md) — Safe read-only commands, source-level redaction checks, non-Git/archive/OCI/LFS/multi-license rules, and sandbox isolation rules
  64. - [`references/examples.md`](references/examples.md) — Calibrated Light & Deep output examples plus common anti-patterns
  65. - [`assets/`](assets/) — Repository hero banner ([`hero.png`](assets/hero.png), 1280×640, GitHub Social Preview ready) and square icon ([`logo.png`](assets/logo.png), 512×512)
  66. - [`README.md`](README.md) / [`README.zh-CN.md`](README.zh-CN.md) — English and Simplified Chinese documentation
  67. - [`CHANGELOG.md`](CHANGELOG.md) — Release history
  68. - [`LICENSE`](LICENSE) — MIT License

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.