The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the configuration of arbitrary shell commands via the custom_cli_command and openclaw.gateways settings. It provides templates that execute shell commands (e.g., clawdbot agent) with redirections (>>, 2>&1) and logic gates (|| true).
[DATA_EXFILTRATION]: The configuration allows the setup of custom_webhook_command and various native webhooks (Discord, Slack). This establishes a mechanism to send session data, instructions, and questions to external, user-defined URLs.
[CREDENTIALS_UNSAFE]: The skill is designed to collect and store sensitive platform tokens and webhook URLs in a local configuration file at ~/.codex/.omx-config.json.
[PROMPT_INJECTION]: The skill implements an indirect prompt injection surface by interpolating dynamic event data into command and webhook templates.
Mandatory Evidence Chain (Category 8):
Ingestion points: Untrusted data enters the context via variables like {{instruction}}, {{question}}, {{reason}}, and {{event}} in SKILL.md.
Boundary markers: None present; variables are interpolated directly into shell command strings and notification bodies.
Capability inventory: The skill configures shell command execution (custom_cli_command) and network requests (custom_webhook_command).
Sanitization: No sanitization or validation logic is provided within the skill's instructions for the interpolated content.
[EXTERNAL_DOWNLOADS]: The verification guidance includes running npm run build, which typically triggers the execution of build scripts and the retrieval of external dependencies from public registries.

configure-notifications