The preparatory work consists of two parts: first, generating the ccxml corresponding to the AWR2944P family target; second, generating a JS script that covers the complete action chain (Connect Core, CPU Reset, Load Memory, Halt CPU, Read Memory, Save Memory).

The ccxml can be automatically generated by CCS according to the target device configuration, so it will not be elaborated here.

The JS script is not created out of thin air; rather, it is structurally composed and parametrically extended based on TI's official scripting patterns. Specifically, the Agent first studies the example scripts in the CCS scripting directory, extracts stable and reusable API invocation paradigms and exception-handling practices, and then organizes the actions into an executable state-machine workflow according to the user's objectives. On this basis, the script needs to be explicitly parameterized, including the ccxml path, image path, read address, read length, result save path, and timeout threshold, so that it can be reused across different projects, different images, and different test rounds without repeatedly modifying the main logic. A JS script that covers the complete debugging action chain. The complete action chain should include at least six steps—Connect Core, CPU Reset, Load Memory, Halt CPU, Read Memory, and Save Memory—organized in causal order rather than simply concatenating commands. Figure 3-1 shows the JS script generated by the Agent, which can implement the simple function of Load Image. All interfaces use APIs from the example JS scripts, combined for specific scenarios.

As demonstrated, the AI Agent can learn how these APIs work by referring to the JS examples provided by CCS, then combine them as needed to meet user requirements and automate the repetitive task of writing and rewriting scripts.