The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/redis_crud.sh executes system binaries such as redis-cli and ssh to manage database connections. It also utilizes network utilities (nc, lsof, ss, netstat) to check for available ports when establishing SSH tunnels.
[REMOTE_CODE_EXECUTION]: In ssh-remote mode, the skill executes a Bash script snippet on a remote host via SSH. This is a core feature enabling the agent to interact with Redis instances that are only accessible from specific remote servers.
[DATA_EXFILTRATION]: The skill includes functionality to read remote configuration files (e.g., .env) to identify Redis connection strings. This allows the agent to utilize existing credentials on a server without requiring the user to provide them manually.
[PROMPT_INJECTION]: Indirect Prompt Injection Surface Analysis:
Ingestion point: Redis command results captured in scripts/redis_crud.sh from redis-cli output.
Capability inventory: Subprocess execution (ssh, redis-cli) and local file writes to the configuration directory.
Sanitization: Employs a json_escape function to sanitize data before interpolation into the JSON response.
Boundary markers: The skill outputs structured JSON, providing clear boundaries between metadata and retrieved Redis data.
[COMMAND_EXECUTION]: The script dynamically loads profile configurations from the user's home directory using the source command. The implementation restricts profile filenames to an alphanumeric character set and enforces file permissions (chmod 600) to protect the configuration files.

redis-crud