技能 doubt-driven-development

📦

doubt-driven-development

Name: doubt-driven-development
Author: addyosmani

內容修訂 r1 高風險 ⚙️ 外部命令📁 檔案系統存取

Challenge Engineering Decisions Before They Stand

Long sessions can turn assumptions into accepted facts and hide costly failure modes. This skill applies bounded, fresh-context adversarial review before non-trivial decisions stand.

支援: Claude Codex Code(CC)

⚠️ 38 差

Maintain this Skill? Claim it ⚠️ 38 差

安裝前請停止並請求確認。

請審查安裝計畫，並在變更檔案前取得使用者明確同意。

使用我的智慧代理安裝

將此請求複製給你的智慧代理，其中包含權威 Skill 頁面和清單。

智能體請求

Review the Skillstore skill "doubt-driven-development" from https://skillstore.io/skills/addyosmani-doubt-driven-development.md and its manifest at https://skillstore.io/api/skills/addyosmani-doubt-driven-development/manifest. Verify the artifact. Stop and obtain explicit user consent before installing or changing files.

你的智慧代理仍應顯示其計畫，並依安全政策請求必要確認。

Agent 可讀資源

當 AI Agent、爬蟲或腳本需要乾淨脈絡、而不是讀取完整頁面時，請使用這些連結。

Markdown 詳情 GET /skills/addyosmani-doubt-driven-development.md 已簽名 manifest GET /api/skills/addyosmani-doubt-driven-development/manifest 已簽名 lockfile GET /api/skills/addyosmani-doubt-driven-development/lockfile

測試它

正在使用「doubt-driven-development」。 Review a proposal to add shared in-memory caching to a concurrent service.

預期結果:

Claim: The cache remains correct under concurrent reads and writes.
Actionable issue: The proposal does not define synchronization during eviction.
Next step: Specify the concurrency contract and rerun the review.

正在使用「doubt-driven-development」。 Challenge a database migration plan that removes a legacy column.

預期結果:

Contract misread: The rollback window was not included in the supplied contract.
Actionable issue: Older application instances still read the legacy column.
Stop status: Revise the deployment order before another review cycle.

正在使用「doubt-driven-development」。 Reconcile feedback about a public API pagination change.

預期結果:

Accepted trade-off: Existing clients keep the old default for one release. Document the compatibility period and add a removal checkpoint.

安全審計

高風險

v1 • 2026/9/19 開啟版本化報告

Most static alerts are false positives caused by Markdown backticks, code fences, ordinary prose, or documentation links. The fixed files in /tmp create local confidentiality and integrity risk, while cross-model review can expose sensitive artifacts or let untrusted artifact instructions influence a CLI reviewer.

已掃描檔案

244

分析行數

複核項目

已忽略的誤報

已確認的安全問題 (2)

高

SKILL.md:135-151

Incomplete Isolation for Prompt-Injected Artifacts

The skill recognizes that artifacts may contain prompt injection but treats a read-only sandbox as sufficient containment. A reviewer launched inside the repository may still read sensitive files or use network capabilities, depending on the CLI.

The examples launch model CLIs with the repository as context, and the text explicitly anticipates prompt injection. Read-only access protects writes but does not universally prevent reads or outbound disclosure.

中

SKILL.md:120-132

Sensitive Artifact Disclosure to External Models

The workflow offers to send artifacts and contracts to external model CLIs without requiring a secrets, ownership, or data-retention check. User authorization for execution does not establish that the content may leave its trust boundary.

The text explicitly offers Gemini or Codex review and specifies which artifact content to submit. It discusses authorization and credentials but not content sensitivity or provider policy.

能力複核項目 (2)

這些是真實的本機能力，對此技能而言可能是預期行為，因此需要複核，但不會按已確認的惡意行為計分。

中

SKILL.md:144

Temp directory access

codex exec --sandbox read-only -C <repo-path> - < /tmp/doubt-prompt.md

The example reads a predictable shared /tmp filename that may contain proprietary code or secrets. Another local process could replace or read that file unless creation, permissions, and cleanup are secured.

中

SKILL.md:148

Temp directory access

gemini --approval-mode plan -p "" < /tmp/doubt-prompt.md

The Gemini example also consumes the fixed /tmp/doubt-prompt.md path. Predictable shared temporary files create local tampering and disclosure risk.

審計單位: codex

分享與引用此報告

分享版本化評估報告、中性徽章、嵌入卡片與引用。Skillstore 只報告證據，不判斷該 Skill 是否安全。

開啟版本化報告

複製報告連結

https://skillstore.io/skills/addyosmani-doubt-driven-development/audits/1?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Markdown 徽章

[![Skillstore security assessment](https://skillstore.io/badges/skills/addyosmani-doubt-driven-development/security.svg)](https://skillstore.io/skills/addyosmani-doubt-driven-development?utm_source=security_passport_badge)

HTML 徽章

<a href="https://skillstore.io/skills/addyosmani-doubt-driven-development?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/addyosmani-doubt-driven-development/security.svg" alt="Skillstore security assessment" loading="lazy"></a>

嵌入卡片

<iframe src="https://skillstore.io/embed/skills/addyosmani-doubt-driven-development.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>

學術引用 (APA · BibTeX · CFF)

APA 引用

addyosmani. (2026). doubt-driven-development security audit report (audit version 1) [Author version unspecified]. Skillstore. https://skillstore.io/skills/addyosmani-doubt-driven-development/audits/1

BibTeX 引用

@techreport{addyosmani-addyosmani-doubt-driven-development-2026,
  author = {addyosmani},
  title = {doubt-driven-development security audit report (audit version 1)},
  institution = {Skillstore},
  year = {2026},
  number = {1},
  url = {https://skillstore.io/skills/addyosmani-doubt-driven-development/audits/1},
  note = {Author version unspecified}
}

CITATION.cff

cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "doubt-driven-development security audit report (audit version 1)"
version: "unspecified"
type: report
authors:
  - name: "addyosmani"
date-released: "2026-09-19"
url: "https://skillstore.io/skills/addyosmani-doubt-driven-development/audits/1"
identifiers:
  - type: other
    value: "skillstore:addyosmani-doubt-driven-development:audit:1"
    description: "Skillstore immutable audit report identifier"

Skillstore 評分

評分依據證據置信度: 中

架構

可維護性

內容

社群

規範符合度

你可以打造什麼

Validate an Architecture Decision

Challenge service boundaries, shared state, failure modes, and unstated assumptions before implementation begins.

Review Security-Sensitive Logic

Test authentication, authorization, or data-handling changes against an explicit contract before committing.

Prepare a High-Risk Migration

Expose irreversible steps, rollback gaps, and hidden dependencies before approving a production migration.

試試這些提示

Review One Decision

Apply doubt-driven development to this decision. State the claim, extract the artifact and contract, then find assumptions and failure modes.

Challenge a Code Change

Review this change against its contract. Focus on edge cases, hidden state, broken conventions, and inputs that violate expected behavior.

Reconcile Reviewer Findings

Classify each finding as contract misread, actionable issue, accepted trade-off, or noise. Cite artifact evidence and recommend the next step.

Run a Bounded Review Cycle

Run up to three doubt cycles on this high-stakes artifact. Track changed assumptions, unresolved findings, accepted trade-offs, and the final stop condition.

最佳實務

Keep each artifact small and provide a precise contract.
Use reviewer findings as evidence, then verify each finding against the artifact.
Record the stop condition and escalate unresolved issues after three cycles.

避免

Do not pass the original claim when independent contract evaluation is required.
Do not invoke external model tools without explicit authorization for each call.
Do not repeat reviews on an unchanged artifact to manufacture confidence.

常見問題

How is this different from a final code review?

It challenges non-trivial decisions during development, while a final review evaluates a completed artifact.

What counts as a non-trivial decision?

Examples include branching logic, module boundaries, unverifiable invariants, irreversible changes, and decisions with hidden context.

Does every change need a doubt cycle?

No. Mechanical edits, formatting, simple file moves, and obvious one-line changes usually do not need one.

Must I use another model?

No. Interactive sessions should offer cross-model review, but the user may skip it.

When should the review loop stop?

Stop after trivial findings, three cycles, or an explicit user decision to proceed.

Can reviewer findings be accepted automatically?

No. Recheck every finding against the artifact and classify it before changing the work.

開發者詳細資訊

作者

addyosmani

授權

MIT

Skillstore 修訂

版本提示

作者未聲明版本。

儲存庫

https://github.com/addyosmani/agent-skills/tree/c004a74784a08295d52749b04cda634125b9a581/skills/doubt-driven-development

引用

5d5054f8a23586f9b500fece1cb613a9dffc787b

維護新鮮度

2026/9/19

使用量

0 次下載 · 0 次瀏覽

檔案結構

📄 SKILL.md

doubt-driven-development

使用我的智慧代理安裝

Agent 可讀資源

測試它

安全審計

已確認的安全問題 (2)

風險因素

複製報告連結

Markdown 徽章

HTML 徽章

嵌入卡片

APA 引用

BibTeX 引用

CITATION.cff

Skillstore 評分

你可以打造什麼

Validate an Architecture Decision

Review Security-Sensitive Logic

Prepare a High-Risk Migration

試試這些提示

最佳實務

避免

常見問題

開發者詳細資訊