The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs shell commands using raw user-provided text without sanitization. Specifically, the instruction deepline feedback send "{feedback text}" allows for shell command injection if the user provides input containing metacharacters like semicolons, backticks, or command substitutions.
[DATA_EXFILTRATION]: The skill is designed to exfiltrate current session transcripts (in .jsonl format) to the vendor's domain. While it asks for user consent for requested reports, these files contain the full interaction history and may include sensitive code or data from the environment.
[EXTERNAL_DOWNLOADS]: The skill performs an automated installation of a global NPM package (deepline) from a non-standard registry (https://code.deepline.com/api/v2/npm/).
[INDIRECT_PROMPT_INJECTION]: By reading and transmitting session history which contains untrusted external data (such as web search results or file content processed during the session), the skill creates a surface where malicious content in the history could influence the reporting process or lead to unauthorized data transmission.
[DATA_EXPOSURE]: The skill generates "proactive" diagnostic reports without user confirmation. While these reports are intended for performance monitoring, they transmit metadata such as session goals and friction points which may contain sensitive context.

deepline-feedback