The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts to manage the approval workflow and competitor tracking.
Evidence: Shell command execution of python "${CLAUDE_PLUGIN_ROOT}/scripts/approval-manager.py" for creating and updating execution approvals.
Evidence: Execution of competitor-tracker.py for processing audience segment data.
[INDIRECT_PROMPT_INJECTION]: The skill processes structured data from external, potentially untrusted sources which creates an attack surface for indirect prompt injection.
Ingestion points: The skill reads raw data from CSV files, JSON files, and remote Google Sheets URLs provided by the user.
Boundary markers: While there is a mandatory human approval gate (yes/no) before execution, the instructions lack clear delimiters or specific warnings to ignore instructions embedded within the imported data fields.
Capability inventory: The skill has the ability to write data to external CRM/Email platforms and execute shell commands via the approval manager.
Sanitization: The process includes format validation (email RFC 5322) and schema inference, but does not explicitly describe sanitization techniques to prevent the agent from interpreting data content as instructions.
[DATA_EXFILTRATION]: The skill accesses sensitive brand configuration and internal documentation stored in hidden directories within the user's home folder.
Sensitive file paths: Reads from ~/.claude-marketing/brands/_active-brand.json, ~/.claude-marketing/brands/{slug}/profile.json, and ~/.claude-marketing/sops/.
Transfer point: Data from these files is used to influence compliance logic and import parameters, which are subsequently sent to external destination platforms (e.g., HubSpot, Salesforce, Mailchimp).

data-import