The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/manage-humans-registry.sh is vulnerable to JQ injection in the action_update function. It constructs a jq filter string by directly interpolating user-controlled variables (NAME, LEVEL, TEAMS, WORKERS) without sanitization. This allows an attacker to manipulate the structure and contents of the humans-registry.json file through crafted inputs.
[CREDENTIALS_UNSAFE]: The scripts/create-human.sh script generates a random password for new users and outputs it in plain text within the final JSON result block. This causes the credential to be recorded in the agent's conversation history and system logs, making it accessible to anyone with access to those logs.
[COMMAND_EXECUTION]: The email notification logic in scripts/create-human.sh is potentially vulnerable to mail header injection. User-provided fields like DISPLAY_NAME are interpolated into the email body and passed to msmtp via echo -e, which could allow an attacker to inject additional SMTP headers such as Bcc or Cc if the input contains newline characters.
[CREDENTIALS_UNSAFE]: The skill accesses and modifies sensitive system configuration files such as /data/hiclaw-secrets.env and agent openclaw.json files. While these operations are part of the skill's administrative functionality, they represent a significant attack surface involving credentials and configuration secrets.

human-management