The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the b2c CLI tool (or @salesforce/b2c-cli via npx) to perform debugging tasks. This includes starting interactive sessions (b2c debug cli), inspecting configuration (b2c setup inspect), and running a JSONL-based RPC interface for programmatic control.
[DYNAMIC_EXECUTION]: The debugger provides an eval command (and an evaluate RPC command) that allows the execution of arbitrary expressions within the runtime context of the B2C Commerce script being debugged. This is standard functionality for a debugger but constitutes a dynamic execution surface.
[CREDENTIALS_UNSAFE]: The skill documentation describes the use of b2c setup inspect --unmask to reveal resolved configuration secrets. It also references standard locations where the CLI searches for credentials, including dw.json, .env, ~/.mobify, and package.json files.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external servers (Salesforce B2C Commerce instances) that could be controlled by an attacker if the script environment is compromised.
Ingestion points: The agent receives variable values, stack traces, and expression evaluation results from the remote instance via the CLI or MCP diagnostics tools.
Boundary markers: The instructions do not provide explicit boundary markers or instructions to the agent to disregard instructions embedded within the retrieved script state.
Capability inventory: The skill allows for local command execution, file system interaction (via source mapping), and network communication with commerce instances.
Sanitization: There is no evidence of sanitization or filtering applied to the data retrieved from the debugger before it is presented to the agent.

b2c-debug