The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts (campaign-tracker.py, execution-tracker.py, and report-generator.py) to automate data collection and report formatting. These scripts use command-line arguments derived from user input and local configurations.
[DATA_EXFILTRATION]: The skill is designed to read sensitive brand profiles, compliance guidelines, and agency SOPs from the ~/.claude-marketing/ directory and transmit summarized findings to external services including Slack, Email, and Google Sheets. This behavior is mitigated by a required explicit approval step before any external delivery.
[INDIRECT_PROMPT_INJECTION]: The skill ingests performance metrics and brand context from external MCP servers and local configuration files, which could potentially contain adversarial instructions designed to influence the agent's analysis or report narrative.
Ingestion points: Brand profile metadata and connected MCP performance logs (SKILL.md Steps 1 and 2).
Boundary markers: A mandatory "approval checkpoint" (Step 11) serves as a human-in-the-loop boundary to review all generated content before it is finalized or delivered.
Capability inventory: Execution of local Python scripts for data processing; write access to external messaging and document platforms via MCP integrations.
Sanitization: No explicit sanitization or filtering of external data is described prior to interpolation into the reporting prompts.

client-report