The Agent Skills Directory

[CREDENTIALS_UNSAFE] (HIGH): The skill explicitly prompts the user for a Client Secret and a Database URL (which typically contains a username and password). These secrets are then written in plain text to .secondme/state.json on the local filesystem.
[DATA_EXPOSURE] (HIGH): The workflow includes reading and displaying the contents of .secondme/state.json, which contains sensitive credentials. If this skill is used in a shared or logged environment, these secrets could be exposed in the conversation history.
[PROMPT_INJECTION] (LOW): The skill ingests untrusted user input (App Info and App Name) and interpolates it into CLAUDE.md. This represents an indirect prompt injection surface where a malicious payload in the metadata could influence an agent's behavior when it later reads the project documentation.
Ingestion points: User-provided App Info text and manual input fields.
Boundary markers: None identified for the generated CLAUDE.md file.
Capability inventory: File system write operations for configuration and documentation.
Sanitization: None identified; user input is directly mapped to the configuration and documentation fields.

secondme-init