The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions in SKILL.md and references/lldb-patterns.md guide the agent to ingest potentially untrusted data from an application's memory and logs using LLDB commands such as 'v', 'po', and 'frame variable'. This data could contain malicious instructions designed to influence the agent's behavior. The skill also grants the agent capabilities to modify process state (via 'expr' and 'thread return') and perform file operations (via 'xctrace'), which could be exploited in an injection chain. No boundary markers or sanitization procedures are provided to mitigate this risk.\n- [DYNAMIC_EXECUTION]: The skill promotes the use of dynamic code evaluation through LLDB's 'po' and 'expr' commands, which run Swift and Objective-C code at runtime. Additionally, the references/lldb-patterns.md file explains how to integrate and execute custom Python scripts within the debugger, providing an additional layer of dynamic execution capability.\n- [EXTERNAL_DOWNLOADS]: SKILL.md and references/instruments-guide.md include multiple reference links to 'sosumi.ai'. This domain is an unofficial mirror for Apple documentation and is not recognized as a trusted organization or well-known service.\n- [PERSISTENCE]: The references/lldb-patterns.md guide details how to use the '~/.lldbinit' file to store and automatically load custom commands, aliases, and Python scripts. While standard for developers, this represents a documented mechanism for establishing persistence within the user's debugging environment.

debugging-instruments