📦

監査履歴

ship - 1 監査

2026年7月22日 01:41

The 63 static findings are false positives caused by Markdown backticks, a safe fixed Git command substitution, a protective secret-name check, and a text-entropy heuristic. Semantic review found that the skill can publish and merge changes without a final confirmation gate, uses incomplete sensitive-file protection, and does not specify safe shell argument handling.

スキャンされたファイル

276

解析済み行数

レビュー項目

誤検知を無視

確認済みのセキュリティ上の懸念 (3)

高

SKILL.md:9 SKILL.md:149-154

Automatic Remote Merge Without Confirmation Gate

The workflow directs the agent to run the full commit, push, pull request, and merge pipeline after parsing arguments. It does not require a final user confirmation before publishing changes or merging a pull request.

The skill explicitly instructs a full pipeline and documents push, PR creation, and merge actions, but no final confirmation checkpoint is described before those external changes.

高

SKILL.md:160-169

Incomplete Sensitive-File Protection

The sensitive-file check is limited to selected names in git status, followed by git add -A. This can miss secrets in files with other names or existing tracked sensitive files.

The documented filename check is narrow and the subsequent add-all command stages every change. The skill provides no content-based secret scanning or comprehensive ignore validation.

中

SKILL.md:29-37 SKILL.md:149-154

Unspecified Shell Escaping for Commit Messages and Branch Names

The workflow inserts a user-derived commit message and branch placeholders into Git commands without stating quoting or argument-array requirements. An implementation that builds shell strings can allow command injection.

The source identifies user-supplied arguments and later uses message and branch placeholders in shell-command examples, but it does not prescribe safe command construction.

リスク要因

⚙️ 外部コマンド (50)

監査者: claude

← スキルに戻る