", "root_cause": "", "reasoning_chain": ["", "", ""], "files_examined": ["", ""], "next_steps": "" } ``` ### Step 2: Run Report Generator (USE run_in_terminal TOOL) **You MUST use the `run_in_terminal` tool to execute:** ```bash python3 \/lan\/fed\/etpv\/scripts\/ayush_scripts\/modus_cli_agent\/gemini-cli\/.gemini\/skills\/testcase-analysis\/scripts\/generate_report.py \/lan\/fed\/etpv\/scripts\/ayush_scripts\/modus_cli_agent\/gemini-cli\/.gemini\/skills\/testcase-analysis\/assets\/analysis_.json \/lan\/fed\/etpv\/scripts\/ayush_scripts\/modus_cli_agent\/gemini-cli\/.gemini\/skills\/testcase-analysis\/assets\/ ``` **Expected output:** `✅ Report generated: ...\/assets\/report__.html` ### Step 3: Confirm to User After the script runs successfully, tell the user: ``` 📄 **HTML Report Generated** - Report: `assets\/report__.html` - Open in browser to view formatted analysis ``` ### EXAMPLE - Complete Phase 10 Execution: **For a testcase named `Broadcast_Scan`:** 1. **Create JSON** (use create_file tool): - Path: `...\/assets\/analysis_Broadcast_Scan.json` 2. **Run script** (use run_in_terminal tool): ```bash python3 ...\/scripts\/generate_report.py ...\/assets\/analysis_Broadcast_Scan.json ...\/assets\/ ``` 3. **Confirm**: Tell user the HTML report location --- ## Global References For general operations: - **Rerunning testcases**: [references\/global\/running-testcase.md](references\/global\/running-testcase.md) - **Invoking Modus**: [references\/global\/running_modus.md](references\/global\/running_modus.md) --- ## Critical Principles: The Senior Engineer's Code ### Core Principles 1. **Learn first, analyze second** - Read the infrastructure docs before diving deep into failures 2. **Understand the phases** - Know which of the 10 execution phases failed 3. **Investigate thoroughly** - Check multiple files; don't rely on single data points 4. **Reason, don't pattern-match** - Think about WHY something happened 5. **Prove your conclusions** - Every categorization needs concrete evidence 6. **Eliminate alternatives** - Explain why the failure is NOT the other categories 7. **Be precise** - Cite specific files, line numbers, content 8. **Think like a senior engineer** - Consider context, impact, and edge cases 9. **ALWAYS reflect at checkpoints** - Complete ALL reflection checkpoints; iterate if ANY gaps exist 10. **Never guess** - If confidence is below 90%, go back and gather more evidence before finalizing ### 🔥 Deep Traversal Principles 11. **Follow EVERY trail** - If a path is mentioned, GO CHECK IT. If a script is invoked, READ IT. 12. **Go to the source** - Don't stop at symptoms. Trace to the ROOT CAUSE. 13. **Investigate external paths** - If `\/grid\/common\/pkgs\/python` fails, GO INSIDE and check what's there 14. **Read the scripts** - Makefile calls a script? READ THAT SCRIPT. Script sources another? READ THAT TOO. 15. **Verify infrastructure** - Setup issues might be in infrastructure scripts, not the testcase itself 16. **Use your tools** - You have `ls`, `cat`, `read_file`, terminal access. USE THEM to explore. 17. **Take your time** - Complex debugging takes time. Don't rush to a conclusion. ### 🧠 Extended Thinking Principles 18. **⚠️ MANDATORY: Use activate_deep_reasoning at ALL checkpoints** - You MUST call the tool before critical analysis 19. **Use deep thinking at critical points** - Enable high reasoning effort for root cause analysis and categorization 20. **Think out loud** - Explicitly reason through your evidence and conclusions 21. **Connect the dots** - Use extended thinking to find non-obvious connections between evidence 22. **Challenge yourself** - Use deep reasoning to argue against your own conclusions 23. **Don't rush decisions** - At decision points, engage extended thinking to ensure thoroughness ### 🔧 Using the activate_deep_reasoning Tool **YOU MUST call this tool at the following checkpoints:** 1. **Checkpoint #3: Trail Following Complete** - Before connecting evidence from multiple files 2. **Checkpoint #4: Root Cause Understood** - Before finalizing root cause (if applicable) 3. **DECISION POINT** - Before selecting final category (ALWAYS REQUIRED) 4. **Final Reflection** - Before outputting final analysis (ALWAYS REQUIRED) **Tool Schema:** ```typescript activate_deep_reasoning({ checkpoint: string, \/\/ Which checkpoint you're at effort: "low" | "medium" | "high", \/\/ How much reasoning is needed reason: string \/\/ WHY you need deep thinking at this point }) ``` **Example usage:** ``` activate_deep_reasoning({ checkpoint: "Checkpoint #3: Trail Following Complete", effort: "high", reason: "Need to connect all evidence from multiple files (Makefile, logs, diffs, scripts, paths) to identify root cause" }) ``` **⚠️ FAILURE TO CALL THIS TOOL AT REQUIRED CHECKPOINTS WILL RESULT IN INCOMPLETE ANALYSIS.** ### Reasoning Effort Guidelines | Phase | Reasoning Effort | Rationale | |-------|------------------|-----------| | Phase 1-2: Data gathering | N\/A | No tool call needed | | Phase 3: Status check | N\/A | No tool call needed | | Phase 4: Deep investigation → **Checkpoint #3** | **`high`** | **MANDATORY TOOL CALL** | | Phase 5: Root cause reasoning → **Checkpoint #4** | **`high`** | **MANDATORY if applicable** | | Phase 6: Failure-type analysis | N\/A | Reference-guided | | Phase 7: **CATEGORIZATION → DECISION POINT** | **`high`** | **MANDATORY TOOL CALL** | | Phase 8: Script validation | N\/A | Comparison task | | Phase 9: **FINAL REFLECTION** | **`high`** | **MANDATORY TOOL CALL** | **🧠 You MUST use the activate_deep_reasoning tool at:** - ✅ Checkpoint #3 (Trail following complete) - ✅ Checkpoint #4 (Root cause understood) - if applicable - ✅ DECISION POINT (Category selection) - ALWAYS - ✅ Final Reflection - ALWAYS ### The Senior Engineer's Questions Before finalizing, ask yourself: - "Would a senior DFT PV engineer with 15 years experience accept this analysis?" - "Have I followed every trail to its end?" - "Can I explain the root cause in one clear sentence?" - "If I'm wrong, what's the impact? Have I done enough to be sure?" - "Is there any file I haven't checked that could change my conclusion?" - "Have I used deep thinking at the critical decision points?" ``` ┌─────────────────────────────────────────────────────────────────────────────┐ │ THE SENIOR ENGINEER'S STANDARD │ ├─────────────────────────────────────────────────────────────────────────────┤ │ │ │ A junior engineer reports: "The test failed because file X wasn't found" │ │ │ │ A senior engineer reports: "The test failed because file X wasn't found │ │ at path \/a\/b\/c. This path was set in Makefile line 42, which includes │ │ common.mk from \/infra\/makefiles\/. The variable TOOL_PATH is set from │ │ environment variable $MODUS_HOME, which was \/old\/path instead of │ │ \/new\/path because the setup script \/scripts\/setup.sh line 15 uses a │ │ hardcoded value that doesn't match this machine's installation. │ │ ROOT CAUSE: Infrastructure setup script needs update for new install path. │ │ CATEGORY: SETUP ISSUE" │ │ │ │ BE THE SENIOR ENGINEER. │ │ │ └─────────────────────────────────────────────────────────────────────────────┘ ``` --- ## Configuration Notes ### Enabling Extended Thinking Mode Extended thinking\/reasoning is now supported via the LiteLLM\/JEDAI configuration. To enable it: **Option 1: Via settings.json (Recommended)** Add to your settings to enable reasoning for agents: ```json { "agents": { "overrides": { "codebase_investigator": { "modelConfig": { "generateContentConfig": { "thinkingConfig": { "thinkingBudget": 4096, "reasoningEffort": "high" } } } } } } } ``` **Option 2: Environment Variable** Set the default model to one that supports reasoning: ```bash export LITELLM_DEFAULT_MODEL=GCP_claude-sonnet-4-5 ``` ### Supported Models for Extended Thinking | Model | Reasoning Support | Notes | |-------|-------------------|-------| | GCP_claude-sonnet-4-5 | ✅ Yes | Best for complex analysis | | GCP_claude-opus-4-1 | ✅ Yes | Most capable | | GCP_claude-sonnet-4 | ✅ Yes | Good balance | | GCP_gemini-2.5-pro | ✅ Yes | Fast with good reasoning | | GCP_gemini-2.5-flash | ✅ Yes | Fast | | GCP_qwen\/qwen3-* | ✅ Yes | Good for code | | on_prem_nvidia\/nemotron-* | ✅ Yes | On-premise option | | on_prem_openai\/gpt-oss-120b | ✅ Yes | On-premise option | | AzureOpenAI_gpt-5 | ✅ Yes | Azure option | | AzureOpenAI_o4-mini | ✅ Yes | Azure option | | GCP_claude-3-5-haiku | ❌ No | Does not support reasoning | | GCP_claude-3-5-sonnet-v2 | ❌ No | Does not support reasoning | ### Recommended Reasoning Budgets | Use Case | Budget (tokens) | Reasoning Effort | |----------|-----------------|------------------| | Simple decisions | 512-1024 | low | | Root cause analysis | 2048-4096 | medium-high | | Critical categorization | 4096-8192 | high | | Complex multi-step debugging | 8192-16384 | high | ### JEDAI API Direct Usage For direct API calls with reasoning: **Gemini:** ```python body = { "messages": [...], "model": "GEMINI", "deployment": "gemini-2.5-pro", "reasoning_effort": "high" } ``` **Claude:** ```python body = { "messages": [...], "model": "Claude", "deployment": "claude-sonnet-4", "reasoning_effort": "high" } ``` **Response includes:** ```python response['choices'][0]['message']['content'] # Main response response['choices'][0]['message']['reasoning_content'] # Reasoning trace ``` --- ## FINAL REMINDER: Before You Finish ``` +-----------------------------------------------------------------------------+ | CHECKLIST BEFORE COMPLETING | +-----------------------------------------------------------------------------+ | | | [ ] Did you categorize the testcase? (REGOLD\/SETUP ISSUE\/REPORT TO RnD) | | [ ] Did you provide root cause with evidence? | | [ ] Did you create JSON file in assets\/ using create_file tool? | | [ ] Did you run generate_report.py using run_in_terminal tool? | | [ ] Did you confirm HTML report was generated? | | | | If ANY checkbox is unchecked, GO BACK AND COMPLETE IT. | | | +-----------------------------------------------------------------------------+ ``` **Your analysis is ONLY complete when the HTML report exists in assets\/.**