The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's workflow involves executing multiple local Python scripts (language-router.py, eval-runner.py, brand-voice-scorer.py) that are not included within the skill package, creating a dependency on external, unverified executables.
[COMMAND_EXECUTION]: In Step 4, the variable {translated_content} is directly interpolated into a command-line string: brand-voice-scorer.py --brand {slug} --text "{translated_content}". Since this input can be sourced from arbitrary URLs or user-provided text, it is highly susceptible to shell command injection (e.g., using characters like ;, &&, or |).
[DATA_EXFILTRATION]: The skill accepts URLs as input for both 'Original content' and 'Translated content'. This allows the agent to make outbound network requests to attacker-controlled servers, which could be used for Server-Side Request Forgery (SSRF) or to leak internal data.
[DATA_EXFILTRATION]: The process explicitly reads from several paths in the user's home directory (~/.claude-marketing/), including _active-brand.json, profile.json, _manifest.json, and agency SOPs. Accessing hidden or application-specific directories in the home folder is a characteristic of sensitive data exposure.
[REMOTE_CODE_EXECUTION]: The combination of fetching data from remote URLs and then using that data in a shell command interpolation ({translated_content}) facilitates a remote code execution vector where an external source can dictate the commands run on the host system.

multilingual-score