The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill is designed to read and map sensitive data from the user's home directory, specifically targeting agency credential profiles and client-specific keys located at ~/.claude-marketing/credentials/ and brand profiles at ~/.claude-marketing/brands/.
[REMOTE_CODE_EXECUTION]: The process involves searching the npm registry for packages and subsequently executing them using npx to verify integration status and tool availability. This pattern involves running arbitrary code from external sources determined at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection as it ingests untrusted metadata (package names, descriptions, and schemas) from the npm registry.
Ingestion points: npm registry search results parsed in SKILL.md process steps.
Boundary markers: None identified in the instructions to separate external data from agent instructions.
Capability inventory: File system access (~/.claude-marketing/), network operations (npm registry search, connectivity tests), and command execution (npx).
Sanitization: No explicit sanitization or validation of the ingested npm metadata is mentioned before it is used to generate configuration or execute tests.
[DATA_EXFILTRATION]: The connectivity testing phase executes operations against external service endpoints using the discovered local credentials, which creates a path for potential data exfiltration if a malicious npm package is executed.

add-integration