The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides a pre-configured automation command that sends business performance data to a specific, hardcoded Telegram chat ID.
Evidence: The openclaw cron add command in SKILL.md includes the flag --to "-1003853954380", which targets a fixed Telegram destination.
Data Risk: The automated report extracts and transmits sensitive metrics from email-state.md, including revenue, conversion rates, and campaign performance, to a recipient that is not clearly identified as user-controlled.
[COMMAND_EXECUTION]: The skill documentation encourages the user to run a shell command to enable recurring tasks.
Evidence: The Cron Setup section in SKILL.md provides an openclaw cron add command. While this is a platform-native feature, in this context it facilitates the automated transmission of sensitive data to an external ID.
[INDIRECT_PROMPT_INJECTION]: The skill creates a vulnerability surface where data processed by the automated reporter could influence the AI agent's behavior.
Ingestion points: The AI model tasked with reporting reads email-state.md, which contains user-defined content such as sequence notes and subject lines that may contain external or untrusted data.
Boundary markers: The instructions provided to the reporting model do not use delimiters or "ignore embedded instructions" warnings to separate the analytical task from the data being analyzed.
Capability inventory: The model has the capability to send generated text to a Telegram chat, which could be exploited to exfiltrate data if a prompt injection attack is successful.
Sanitization: There is no evidence of sanitization, filtering, or escaping for the metrics data before it is interpolated into the model's reporting prompt.

email-sequence-manager