Most static findings are false positives caused by Markdown examples, CSS snippets, regex methods, JavaScript template literals, and expected local file operations. I confirmed command-execution risk in sync-brand-to-tokens.cjs because it imports child_process and uses execSync to run a project-local token generator through the shell. The prompt context helper also needs clearer handling of editable brand text as untrusted data.
The script imports child_process and later uses execSync to run a local token generator. External command execution from a community skill can execute workspace-controlled code.
execSync runs a shell command built from the current working directory and project-local script path. A crafted path or compromised local generator can turn this into arbitrary code execution.
The template literal is passed directly to execSync, so it is an actual shell command string. The command uses mostly fixed arguments, but shell interpretation and project-local script execution remain real risks.
This line builds an ImageMagick shell command with imagePath interpolated into a quoted string. It is printed rather than executed, but a crafted path with quotes or shell metacharacters could become unsafe if copied or run by an agent.
This line prints the generated ImageMagick command containing the resolved image path. It is user guidance rather than direct execution, but the path is not shell-escaped.
This line embeds resolvedPath into a Python command example without shell escaping. It is not executed by the script, but unsafe paths could make the suggested command dangerous if copied.
The context injector extracts text from local brand guidelines and interpolates it into a generated prompt addition. A malicious or compromised guidelines file could introduce instruction-like content unless it is clearly delimited as untrusted data.
The code directly extracts markdown code block content and adds it to a prompt addition. I did not find a malicious payload in the skill files, so this is a design risk rather than a confirmed injection attempt.
04
是正措置
この監査で推奨される修正が記録されています。これらを適用する責任はメンテナーにあります。
FIX-001
高
Shell execution during token sync
Replace execSync string commands with execFileSync or spawnFileSync using argument arrays. Require confirmation before running project-local generator scripts.
FIX-002
中
Unescaped shell command suggestions
Do not build copy-paste shell commands with raw file paths. Display separate arguments or apply robust shell escaping before presenting commands.
FIX-003
中
Prompt context built from editable brand files
Delimit imported brand text as untrusted data. Filter prompt-control phrases before adding brand content to generated instructions.