SPRADU5 October   2026 AWR2944P

 

  1.   1
  2.   Abstract
  3.   Trademarks
  4. 1Basic Principles of AI-Automated Debugging
  5. 2Agent, Skill, and Prompt
  6. 3Verifying the AI Agent
    1. 3.1 Generating the ccxml File and JS Script
    2. 3.2 Test Verification
  7. 4Using the AI Agent to Assist Debugging
    1. 4.1 Simulating Stress Testing
    2. 4.2 Simulating Scenario Testing
  8. 5Conclusion
  9. 6References

Test Verification

The purpose of this test is to verify the reliability of the AI Agent and whether the overall workflow can be fully connected end to end. The Memory address read is 0xA500 0000, which stores ADC data, ensuring the randomness of the test data.

Test method:

  1. Use instructions to have the AI Agent automatically connect to the AWR2944LC and perform reset, image loading, program halt, Memory read, and save.
  2. Disconnect the AI Agent, keep the AWR2944LC EVM board state unchanged, and use CCS to connect to the AWR2944LC and read the memory at the same address.
  3. By comparing whether the Memory obtained by the two methods is identical, it can be determined whether the overall workflow executed by the AI Agent is correct.

Hardware preparation: The AWR2944LC EVM board has already been flashed with the SDK default demo: awr2x44LC_mmw_demoDDM.appimage. After power-on, load the default cfg file to transmit waves, and it can be seen from the mmWave Demo Visualizer that the EVM board is working properly.

 Point cloud results in the wave-transmitting stateFigure 3-2 Point cloud results in the wave-transmitting state

At this point, use the AI Agent to input a prompt, and based on the AWR2944LC ccxml and C:\ti\ccs2040\ccs\scripting as references, generate a Skill for connecting to the AWR2944LC EVM and reading the first 16 words of Memory at 0xA500 0000. The generated Skill is as follows:

 Skill folder generated by the AI AgentFigure 3-3 Skill folder generated by the AI Agent

Including:

  1. Skill.md is used to specify the order in which the Agent executes tasks
  2. Prompt.md is used to specify the default input prompt
  3. JS script, the script generated by the Agent for executing actual tasks
  4. AWR2944LC.ccxml, the configuration file for connecting to the XDS110
  5. Quick Guide, the manual generated by AI on how to use the Skill

 Test invocation workflowFigure 3-4 Test invocation workflow

Using this Skill, the AI Agent can quickly use the CCXML configuration file to connect to the AWR2944EVM and read the specified Memory:

 Agent execution processFigure 3-5 Agent execution process

As shown in Figure 3-4, the upper right corner shows the generated Skill. Based on this Skill, the Agent reads the previously generated JS script and adds the Memory-reading portion to the JS. It can be seen that the Agent executes the JS script, and according to the contents of the JS, the ccxml has already been configured into it, connecting to the MSS and DSS of the AWR2944LC through the XDS110 and halting them.

 APIs actually invoked by the AI AgentFigure 3-6 APIs actually invoked by the AI Agent

The first 16 words of 0xA500 0000 read are as follows:

 The first 16 words of 0xA5000000 read by the AI AgentFigure 3-7 The first 16 words of 0xA5000000 read by the AI Agent

Read the first 32 words at the same address again to verify repeatability:

 The first 16 words of 0xA5000000 read by the AI AgentFigure 3-8 The first 16 words of 0xA5000000 read by the AI Agent

CCS: After using the AI Agent to capture the specified Memory, keep the board state unchanged and directly use CCS to connect to the AWR2944LC EVM, then locate the Memory at the corresponding address. The results are as follows:

 CCS reading 0xA5000000 MemoryFigure 3-9 CCS reading 0xA5000000 Memory

The comparison results are consistent, indicating that the Agent workflow executes correctly and that the read convention is consistent with CCS. After repeated reads with an increased number of words, the results remain consistent, further verifying repeatability and stability.