The Agent Skills Directory

[DATA_EXFILTRATION]: The setup command performs broad scanning of sensitive local files to harvest credentials. It targets shell profiles (~/.zshrc, ~/.bashrc, ~/.zprofile, ~/.envrc) to extract API keys and tokens.
[DATA_EXFILTRATION]: The skill includes code to query the Google Chrome browser history database (~/Library/Application Support/Google/Chrome/Default/History) using sqlite3 to auto-detect marketing platforms and account identities, which is a highly invasive data exposure pattern.
[DATA_EXFILTRATION]: The setup routine attempts to extract secrets from multiple password managers and local stores, including Dashlane (dcli), Doppler CLI, and the macOS Keychain (security find-generic-password).
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to execute curl commands and shell scripts that interpolate numerous environment variables and configuration values, increasing the risk of command injection if these variables are not strictly controlled.
[INDIRECT_PROMPT_INJECTION]: The optimize command spawns a sub-agent (claude-sonnet-4-5) and instructs it to read pre-gathered data from external APIs (Meta, Google Ads, Klaviyo). This creates a vulnerability surface where malicious content embedded in marketing campaigns or API responses could influence the behavior of the sub-agent.
Ingestion points: External API responses from Klaviyo, Meta Graph API, Google Analytics Data API, and Google Search Console API.
Boundary markers: None identified in the instructions for the sub-agent or the data processing scripts.
Capability inventory: Sub-agents have access to the Bash, Agent, and TeamCreate tools.
Sanitization: The skill uses jq to extract values but does not perform content sanitization or validation before passing data to the optimizer agent.

ops-marketing