The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external, untrusted sources (process memory, local variables, thread backtraces) which could contain malicious instructions designed to influence the agent.
Ingestion points: Commands such as memory read, memory find, frame locals, and expr in SKILL.md bring data from a debugged process into the agent context.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat debugged data as untrusted or to ignore embedded instructions.
Capability inventory: The skill allows for process management, file loading (core load), and arbitrary expression evaluation through LLDB (SKILL.md).
Sanitization: There is no documented mechanism for sanitizing or filtering data retrieved from process memory before it is processed by the agent.
[COMMAND_EXECUTION]: The skill provides a set of commands that interact directly with the host system's processes, allowing the agent to launch, attach to, and control arbitrary executables via LLDB.
[DYNAMIC_EXECUTION]: The skill uses the expr command to evaluate expressions within the debugged process's context, which involves runtime evaluation of code snippets via the LLDB Python API.
[EXTERNAL_DOWNLOADS]: The skill instructions include a command to install the cli-anything-lldb package, which is a resource associated with the vendor 'hkuds'.

cli-anything-lldb