The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses the ! syntax to execute a shell command (cat) during initialization. This command reads .agents/product-marketing-context.md to automatically inject business context into the agent's instructions.
[COMMAND_EXECUTION]: The skill frequently invokes a local Python utility script (ds_utils.py) via the shell to process data from various sources. This script handles tasks like campaign analysis, SEO query classification, and Stripe data processing.
[DATA_EXFILTRATION]: The skill accesses high-sensitivity business and customer data, including Stripe subscription details, transaction amounts, churn reasons, and customer email addresses (customer_email), to perform retention analysis. This represents significant data exposure, though no exfiltration to external third-party domains was detected.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests untrusted data from external marketing APIs (Google Ads, GA4, Search Console, Stripe) and allows for manual ingestion of user-provided datasets (CSV, TSV, JSON) in Step 2 Path B.
Ingestion points: Dataslayer MCP (Stripe, GA4, Search Console), Manual user input (CSV/JSON).
Boundary markers: Not explicitly defined in the instructions for data processing.
Capability inventory: The skill has access to the Agent tool for subagent orchestration and can execute shell commands via the Python script invocation.
Sanitization: No specific sanitization or validation steps are outlined for the incoming marketing or financial data.

ds-brain