The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes external system binaries to flash hardware and capture logs.\n
Evidence: subprocess.run() is used within scripts/jlink_flasher.py to call the JLinkExe (or JLink.exe) executable across functions such as check_jlink, detect_connected_devices, run_flash, and run_rtt.\n
Context: The commands are invoked using an argument array without enabling shell=True, which conforms to security best practices and mitigates standard shell command injection risks.\n- [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection or tool output poisoning attack surface by parsing untrusted workspace files.\n
Ingestion points: scripts/jlink_flasher.py reads project-level configuration data from .vscode/launch.json and scans .jlink files within the workspace via scan_jlink_configs.\n
Boundary markers: No boundary markers or clear encapsulation strings are applied to the parsed configurations when they are printed to standard output.\n
Capability inventory: The skill can execute local command-line tools via subprocess.run to interact with physical debugging hardware.\n
Sanitization: Input values retrieved from JSON parsing (such as the device key) are used directly without sanitization or structural validation, meaning a malicious workspace could include instructions disguised as configuration fields to manipulate agent behavior when the tool outputs are read.

flash-jlink