The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands such as python campaign-tracker.py --brand {slug} --action list-campaigns where the {slug} parameter is dynamically loaded from ~/.claude-marketing/brands/_active-brand.json. If the contents of this file are manipulated or if the slug is derived from unsanitized user input, it could lead to arbitrary command execution or argument injection.
[PROMPT_INJECTION]: The skill implements an indirect prompt injection surface by loading instructions, restrictions, and voice-and-tone guidelines from external files like guidelines/_manifest.json and brand profile JSONs. This creates a vulnerability where malicious content within these data files could influence the agent's behavior during the execution of the ROI analysis.
[PROMPT_INJECTION]: Indirect Prompt Injection Evidence Chain:
Ingestion points: ~/.claude-marketing/brands/_active-brand.json, ~/.claude-marketing/brands/{slug}/profile.json, and guidelines/_manifest.json.
Boundary markers: Absent. The skill does not define delimiters to separate data from instructions.
Capability inventory: Subprocess execution (python campaign-tracker.py, scripts/roi-calculator.py) and file system read/write access.
Sanitization: None described. The agent is instructed to directly apply the context and rules found in the files.

roi-calculator