The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill is designed to verify the presence and validity of sensitive files, specifically reaching into ~/.claude-marketing/brands/{brand}/credentials.json and environment variables. Although the instructions state that credentials should not be printed, the agent's access to files containing secrets represents a significant exposure risk.\n- [COMMAND_EXECUTION]: The skill uses Bash to execute shell commands, including curl probes and local Python scripts. It interpolates values from brand configuration files (.mcp.json) directly into shell loops, creating a potential vector for command injection if these files contain malicious payloads.\n- [DATA_EXFILTRATION]: The skill performs network requests using curl to URLs defined in the .mcp.json file. While intended for reachability checks, this capability could be abused to send data to external servers if the configuration is controlled by an attacker.\n- [PROMPT_INJECTION]: The skill ingests untrusted data from profile.json and .mcp.json to drive its validation logic and shell commands, creating a vulnerability surface for indirect injection.\n
Ingestion points: Files located at ~/.claude-marketing/brands/{brand}/profile.json and .mcp.json.\n
Boundary markers: Absent; there are no delimiters or explicit instructions for the agent to ignore embedded instructions in the configuration data.\n
Capability inventory: Read tool for files, Bash tool for script execution, and curl for network connectivity.\n
Sanitization: Absent; the shell interpolation does not include escaping or validation of the extracted URLs from the configuration files.

validate-profile