The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/jlink_debugger.py executes local debugger tools like JLinkGDBServerCLExe and gdb. The implementation uses argument lists rather than shell strings, which mitigates common command injection risks.
[DYNAMIC_EXECUTION]: The skill dynamically generates GDB command files in a temporary directory to automate debugging workflows (e.g., reset, halt, and backtrace). These scripts are executed by GDB in batch mode using the -x flag.
[INDIRECT_PROMPT_INJECTION]: The skill ingests file paths and device names which are interpolated into the generated GDB script, creating a vulnerability surface for command injection within the GDB environment.
Ingestion points: User-controlled --elf and --device arguments in scripts/jlink_debugger.py.
Boundary markers: None; the generated GDB script lacks delimiters or verification to prevent embedded instructions in file paths from being executed as GDB commands.
Capability inventory: The script possesses capabilities to write to the file system and execute subprocesses.
Sanitization: Path separators are normalized to POSIX style, but there is no validation to ensure input strings do not contain newline characters or GDB-specific control sequences.

debug-jlink