The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes scripts/eval-config-manager.py and interpolates user-provided JSON strings ({weights_json}, {overrides_json}) directly into command-line arguments. This presents a vector for command injection if the input contains shell metacharacters such as single quotes, backticks, or semicolons.
[DATA_EXPOSURE]: The skill accesses files located in ~/.claude-marketing/, including brand profiles and guidelines. These files are located in the user's home directory and could contain sensitive business information or proprietary marketing data.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from local brand configuration files which could be manipulated to alter the agent's behavior during evaluation recommendations.
Ingestion points: ~/.claude-marketing/brands/{slug}/profile.json and _manifest.json as specified in the Process section.
Boundary markers: Absent. No mention of delimiters or instructions to ignore nested directives within the JSON data.
Capability inventory: The skill is capable of executing shell commands and modifying configurations through the eval-config-manager.py script.
Sanitization: Absent. The skill instructions do not describe any validation, escaping, or filtering of the content loaded from the configuration files before it is used to generate suggestions or run scripts.

eval-config