📦

監査履歴

architecture-navigator - 6 監査

監査バージョン 6

最新 中リスク

Jun 28, 2026, 10:40 AM

Static analysis flagged many shell, weak-crypto, and reconnaissance patterns. Review found most markdown hits are false positives from code fences, folder names, and examples, but the skill does include a bash scanner that enumerates local project files. No evidence found of prompt injection, network exfiltration, credential access, or malicious intent.

3
スキャンされたファイル
659
解析された行数
8
検出結果
codex
監査者
中リスクの問題 (2)
Local Shell Scanner Enumerates Project Files
The bash scanner runs local commands such as find, basename, wc, and date against a supplied project root. This is legitimate architecture discovery, with quoted path variables and no network calls found, but it still executes shell commands and reads local filesystem structure. Verdict: true positive for local command execution. Confidence: 0.84. Confidence reasoning: executable shell commands are directly present, while semantic review shows bounded architecture-scanning intent.
Skill Instructions Encourage Running Local Commands
The skill documentation tells the assistant to run the architecture scanner and includes ls, find, and grep examples. These commands are relevant to codebase navigation and validation, but they increase operational risk because an AI agent may execute them in a user workspace. Verdict: true positive with legitimate use. Confidence: 0.78. Confidence reasoning: the instructions explicitly request local command execution, but no malicious command, network endpoint, or secret collection was found.
低リスクの問題 (3)
Markdown Examples Trigger Command Pattern Alerts
Many external command detections in SKILL.md and references/architecture-rules.md are markdown code fences, inline paths, or illustrative folder trees. They are not executable by themselves and do not show hidden command injection behavior. Verdict: mostly false positive. Confidence: 0.92. Confidence reasoning: the flagged locations are documentation examples, while the executable behavior is isolated to the scanner script and explicit command examples.
Weak Cryptography Alerts Are Documentation False Positives
The high-severity weak cryptography alerts point to architecture text, styling paths, and examples, not cryptographic functions or algorithms. No evidence found of MD5, SHA1, DES, RC4, or custom cryptographic code in the reviewed files. Verdict: false positive. Confidence: 0.96. Confidence reasoning: line context shows folder descriptions and placement examples rather than security-sensitive code.
Filesystem Alerts Reflect Expected Local Paths
The .claude path and /dev/null redirection are expected for a local skill script and shell output handling. They do not access secrets or hidden user data in the reviewed context. Verdict: low-risk usage. Confidence: 0.90. Confidence reasoning: the paths are visible documentation or standard shell redirection, with no evidence of sensitive file reads.

検出されたパターン

Shell Command Execution in Scanner ScriptAgent-Facing Instructions to Run Shell Commands

監査バージョン 5

安全

Jan 16, 2026, 04:22 PM

Pure documentation and architecture guidance skill. Contains one read-only bash script that scans project directories using standard POSIX utilities (find, wc, basename, date). Script only reads project structure and outputs markdown summary to stdout. No network access, no file writes beyond stdout, no sensitive data handling. All 156 static findings are FALSE POSITIVES - documentation examples and standard file reading operations.

4
スキャンされたファイル
878
解析された行数
1
検出結果
claude
監査者
セキュリティ問題は見つかりませんでした

監査バージョン 4

安全

Jan 16, 2026, 04:22 PM

Pure documentation and architecture guidance skill. Contains one read-only bash script that scans project directories using standard POSIX utilities (find, wc, basename, date). Script only reads project structure and outputs markdown summary to stdout. No network access, no file writes beyond stdout, no sensitive data handling. All 156 static findings are FALSE POSITIVES - documentation examples and standard file reading operations.

4
スキャンされたファイル
878
解析された行数
1
検出結果
claude
監査者
セキュリティ問題は見つかりませんでした

監査バージョン 3

低リスク

Jan 10, 2026, 10:22 AM

Pure documentation and architecture guidance skill with one bash script that scans project directories. The script uses standard POSIX utilities to read project structure and outputs markdown summary. No network access, no sensitive data handling, no file writes beyond stdout.

3
スキャンされたファイル
659
解析された行数
1
検出結果
claude
監査者
低リスクの問題 (1)
Bash script executes external commands
The script uses find, wc, basename, and date commands to scan directory structure. These are standard POSIX utilities used for reading project files and counting lines. The script only reads from the project directory and outputs to stdout.

監査バージョン 2

低リスク

Jan 10, 2026, 10:22 AM

Pure documentation and architecture guidance skill with one bash script that scans project directories. The script uses standard POSIX utilities to read project structure and outputs markdown summary. No network access, no sensitive data handling, no file writes beyond stdout.

3
スキャンされたファイル
659
解析された行数
1
検出結果
claude
監査者
低リスクの問題 (1)
Bash script executes external commands
The script uses find, wc, basename, and date commands to scan directory structure. These are standard POSIX utilities used for reading project files and counting lines. The script only reads from the project directory and outputs to stdout.

監査バージョン 1

低リスク

Jan 10, 2026, 10:22 AM

Pure documentation and architecture guidance skill with one bash script that scans project directories. The script uses standard POSIX utilities to read project structure and outputs markdown summary. No network access, no sensitive data handling, no file writes beyond stdout.

3
スキャンされたファイル
659
解析された行数
1
検出結果
claude
監査者
低リスクの問題 (1)
Bash script executes external commands
The script uses find, wc, basename, and date commands to scan directory structure. These are standard POSIX utilities used for reading project files and counting lines. The script only reads from the project directory and outputs to stdout.