testdino-mcp: Command for Claude Code

Investigate why a tool is not working as expected. This command is for debugging tool behavior — wrong responses, auth failures, missing data, or unexpected errors.

Input: $ARGUMENTS — description of the problem (e.g., "list_testruns returns empty for valid projectId", "health tool says PAT is invalid but it works in curl")

---

Phase 1: Reproduce the Problem

1. Identify the tool — Which tool is affected?
2. Read the tool source — Read the handler file completely
3. Trace the data flow:
   • What args does the handler expect?
   • How does it build the API URL? (check endpoints.ts)
   • What does apiRequestJson do with the response?
   • How does it format the output?

---

Phase 2: Check Each Layer

Work from the outside in:

1. Input layer — Is the tool schema correct? Are required/optional fields right? Would the AI agent send the right args based on the description?
2. Auth layer — Does getApiKey(args) resolve correctly? Is the token being passed in the Authorization header?
3. URL layer — Is the endpoint URL correct? Are query params properly built? Log the URL if needed
4. API layer — Does the API actually return what we expect? Check if the response is wrapped ({ success, data } vs direct)
5. Response layer — Is the response being formatted correctly for MCP? Is JSON.stringify applied?

---

Phase 3: Fix

1. Fix the root cause — don't add workarounds
2. Run npm run typecheck && npm run lint
3. If the fix changes tool behavior, update docs/TOOLS.md

---

Anti-Patterns

Don't	Why
Add try/catch that swallows errors	Hides the real problem
Change the tool description to work around a bug	Fixes the symptom, not the cause
Hardcode values to make it "work"	Breaks for other inputs
Skip checking endpoints.ts	Most URL bugs live there