The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous examples of shell and tool-specific commands (OpenOCD, GDB, J-Link) for flashing firmware, inspecting memory, and controlling processor execution. These commands are essential for firmware debugging but provide the agent with broad system access during debugging sessions. Examples include target connection commands and memory dump operations like dump binary memory /tmp/heap.bin in SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves the agent ingesting and analyzing potentially untrusted data from embedded targets, such as memory dumps and register states. Because the tools used (like GDB) have extensive capabilities including internal scripting, there is a theoretical surface for indirect prompt injection.
Ingestion points: Serial logs, crash dumps, and firmware ELF files analyzed in SKILL.md.
Boundary markers: No explicit instructions are provided to the agent to treat debug data as untrusted or to use specific delimiters.
Capability inventory: Access to GDB (which supports shell commands and Python scripting) and OpenOCD as defined in the mcpConfig and instructions.
Sanitization: The instructions do not include steps to sanitize or validate content before it is processed by debugging utilities.

firmware-debugger