The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill manages sensitive configuration files containing API keys. Specifically, the zread config --stdio interface returns the plaintext llm_api_key in its ViewModel output, exposing stored credentials to any agent or log system processing the tool's machine-readable output.
[EXTERNAL_DOWNLOADS]: The skill documents the installation and update of external binaries from remote sources via npm and the tool's own update command.
[REMOTE_CODE_EXECUTION]: Through the zread update command, the tool can replace its own executable binary with content downloaded from a remote URL at runtime, representing a potential execution vector.
[COMMAND_EXECUTION]: The skill allows the agent to execute multiple CLI commands, including repository crawling (generate), local web serving (browse), and configuration management (config), which involve file system writes and local server hosting.
[DATA_EXFILTRATION]: The core functionality involves sending the local codebase contents to an external LLM provider, which constitutes an inherent data transfer of potentially sensitive intellectual property.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection by processing unvalidated repository content (code and comments) through an LLM to generate wikis.
Ingestion points: Reads all files within the current repository workspace during the generation phase.
Boundary markers: No delimiters or safety instructions are specified in the protocol to separate code content from instructions for the LLM.
Capability inventory: Local file system writes, local server hosting, and self-updating binary capabilities.
Sanitization: No sanitization or filtering of the ingested source code is mentioned in the protocol before it is transmitted to the LLM.

zread