The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a helper script scripts/dlv-session.sh to execute dlv (Delve) and tmux commands. The script correctly uses shell quoting (via printf %q) when building the command string to prevent command injection from user-supplied arguments.
[EXTERNAL_DOWNLOADS]: The documentation recommends installing the Delve debugger from its official repository on GitHub (github.com/go-delve/delve). This is a well-known and trusted source for this developer tool.
[INDIRECT_PROMPT_INJECTION]: The skill captures and displays output from the debugger to the agent. While it is theoretically possible for a debugged program to print malicious instructions intended to trick the agent, this is an inherent risk of any tool that observes external program output and is mitigated by the agent's own safety guardrails.
[DATA_EXPOSURE]: The skill provides the ability to attach to running processes via dlv attach <pid>. This allows the agent to inspect the memory and state of other processes on the system. This functionality is consistent with the skill's stated purpose as a debugger and is necessary for its operation.

go-debug