configure-notifications
Fail
Audited by Gen Agent Trust Hub on Aug 20, 2026
Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONDATA_EXFILTRATIONREMOTE_CODE_EXECUTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill enables the configuration of
custom_cli_commandandopenclawgateways. These features allow for the definition of arbitrary shell command templates that the host system executes automatically in the background when specific events occur, such as session starts or user questions. - [DATA_EXFILTRATION]: Configuration examples provided in the skill include a hardcoded Discord channel ID (
1468539002985644084). If an operator adopts these examples verbatim, sensitive system notifications and session data would be exfiltrated to an external, potentially unauthorized Discord channel. - [CREDENTIALS_UNSAFE]: The skill directly reads and modifies
$HOME/.codex/.rcs-config.json. This configuration file is a known storage location for sensitive authentication tokens, API keys, and private webhook URLs for services like Discord, Telegram, and Slack. - [COMMAND_EXECUTION]: The skill uses
jqandmvto programmatically edit system configuration files. It also instructs the agent to execute shell commands likenpm run buildandtmux list-sessionsfor system verification and troubleshooting. - [INDIRECT_PROMPT_INJECTION]: The skill defines a vulnerability surface by interpolating untrusted runtime data—such as user questions (
{{question}}) and session descriptions ({{instruction}})—directly into shell command templates. - Ingestion points: Runtime variables including
{{instruction}},{{question}},{{reason}}, and{{event}}are ingested from the agent's execution context. - Boundary markers: The skill lacks explicit boundary markers or validation logic to prevent malicious content within these variables from influencing shell execution.
- Capability inventory: The skill configures the
custom_cli_commandandopenclawgateways which possess full shell execution capabilities. - Sanitization: While the skill text claims the underlying platform performs shell-escaping, the skill itself implements no sanitization or verification of the inputs before they are written to the configuration.
Recommendations
- AI detected serious security threats
Audit Metadata