The Agent Skills Directory

[DATA_EXPOSURE]: The skill accesses sensitive local directories used for storing authentication data.
Evidence: Step 1 and Step 7 describe reading from and mapping credentials within ~/.claude-marketing/credentials/.
Evidence: Step 5 instructs the agent to guide users in storing secrets in .env files.
[REMOTE_CODE_EXECUTION]: The skill executes code from external npm packages as part of its testing and configuration workflow.
Evidence: Step 3 generates configurations using npx to run packages directly from the npm registry.
Evidence: Step 6 performs a connectivity test by starting the MCP server and executing a "read-only operation" to verify authentication, which involves running arbitrary code from the third-party package.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources that could contain malicious instructions.
Ingestion points: Step 2 queries the npm registry and external MCP server directories for package metadata (descriptions, tool definitions).
Boundary markers: No specific delimiters or instructions are used to ignore embedded commands in the fetched package metadata.
Capability inventory: The agent has the ability to execute shell commands via npx, read/write local configuration files (.mcp.json), and access sensitive credential paths.
Sanitization: No validation or sanitization of the fetched package metadata is performed before displaying it to the user or incorporating it into configuration files.
[COMMAND_EXECUTION]: The skill constructs and suggests shell commands for execution.
Evidence: Steps 3 and 6 involve the use of npx and other startup commands to initialize and test the integration.

add-integration