The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local script scripts/eval-config-manager.py with arguments derived directly from user input.
Evidence: In SKILL.md, step 4 of the Process section explicitly describes executing scripts/eval-config-manager.py --brand {slug} --action set-weights --weights '{weights_json}' and similar patterns for other actions.
Risk: Passing user-controlled variables like {weights_json} or {overrides_json} inside single quotes to a shell command is vulnerable to argument injection. An attacker could provide a string containing a single quote followed by shell metacharacters (e.g., '; touch /tmp/pwned; ') to achieve arbitrary command execution.
[DATA_EXPOSURE]: The skill reads various configuration files from the user's home directory.
Evidence: SKILL.md mentions loading ~/.claude-marketing/brands/_active-brand.json, ~/.claude-marketing/brands/{slug}/profile.json, ~/.claude-marketing/brands/{slug}/guidelines/_manifest.json, and ~/.claude-marketing/sops/.
Risk: While these paths appear specific to the application, accessing structured data from the home directory constitutes a wide data exposure surface if the skill is manipulated into reading unintended files.

eval-config