The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local openocd and arm-none-eabi-gdb binaries through the Python subprocess module. These operations are necessary for flashing firmware and managing on-chip debugging sessions.
Evidence: scripts/openocd_run.py (line 199), scripts/openocd_gdb.py (line 72), and scripts/openocd_telnet.py (line 49).
[CREDENTIALS_UNSAFE]: The config.json file contains hardcoded absolute file system paths that reveal the author's local Windows username ('甘宇'). While this exposes personally identifiable information (PII), it does not involve sensitive authentication secrets like API keys or passwords.
Evidence: config.json (lines 4 and 7).
[PROMPT_INJECTION]: The skill ingests and parses output from debugger telnet sessions and GDB commands. This creates a surface for indirect prompt injection where malicious data provided by a compromised hardware target (such as register names or memory strings) could be interpolated into the agent's context. The risk is considered low due to the structured parsing used by the scripts.
Evidence: Parsing logic in scripts/openocd_gdb_common.py and response handling in scripts/openocd_telnet.py.

openocd