fix(delegate-task): aggressive tool description to prevent missing category/subagent_type

Problem: Agents frequently omit both 'category' and 'subagent_type' parameters
when calling the task() tool, causing validation failures. The JSON Schema
marks both as optional, and LLMs follow schema structure over description text.

Solution (Option A): Add aggressive visual warnings and failure-mode examples
to the tool description:
- ⚠️ CRITICAL warning header
- COMMON MISTAKE example showing what will FAIL
- CORRECT examples for both category and subagent_type usage
- Clear explanation that ONE must be provided

Tests: All 153 existing tests pass (no behavior change, only prompt improvement)

src/tools/delegate-task/tools.ts

@@ -54,27 +54,45 @@ export function createDelegateTask(options: DelegateTaskToolOptions): ToolDefini
   }).join("\n")
 
   const description = `Spawn agent task with category-based or direct agent selection.
-
+  
-REQUIRED: You MUST provide EITHER category OR subagent_type (one of them is REQUIRED, but not both).
+  ⚠️  CRITICAL: You MUST provide EITHER category OR subagent_type. Omitting BOTH will FAIL.
-- If using a predefined category → provide category
+  
-- If using a specific agent → provide subagent_type
+  **COMMON MISTAKE (DO NOT DO THIS):**
-- Providing NEITHER is INVALID and will fail.
+  \`\`\`
-
+  task(description="...", prompt="...", run_in_background=false)  // ❌ FAILS - missing category AND subagent_type
-- load_skills: ALWAYS REQUIRED. Pass at least one skill name.
+  \`\`\`
-- category: Use predefined category → Spawns Sisyphus-Junior with category config
+  
-  Available categories:
+  **CORRECT - Using category:**
-${categoryList}
+  \`\`\`
-- subagent_type: Use specific agent directly
+  task(category="quick", load_skills=[], description="Fix type error", prompt="...", run_in_background=false)
-- run_in_background: true=async (returns task_id), false=sync (waits for result). Default: false. Use background=true ONLY for parallel exploration with 5+ independent queries.
+  \`\`\`
-- session_id: Existing Task session to continue (from previous task output). Continues agent with FULL CONTEXT PRESERVED - saves tokens, maintains continuity.
+  
-- command: The command that triggered this task (optional, for slash command tracking).
+  **CORRECT - Using subagent_type:**
-
+  \`\`\`
-**WHEN TO USE session_id:**
+  task(subagent_type="explore", load_skills=[], description="Find patterns", prompt="...", run_in_background=true)
-- Task failed/incomplete → session_id with "fix: [specific issue]"
+  \`\`\`
-- Need follow-up on previous result → session_id with additional question
+  
-- Multi-turn conversation with same agent → always session_id instead of new task
+  REQUIRED: Provide ONE of:
-
+  - category: For task delegation (uses Sisyphus-Junior with category-optimized model)
-Prompts MUST be in English.`
+  - subagent_type: For direct agent invocation (explore, librarian, oracle, etc.)
+  
+  **DO NOT provide both.** If category is provided, subagent_type is ignored.
+  
+  - load_skills: ALWAYS REQUIRED. Pass [] if no skills needed, or ["skill-1", "skill-2"] for category tasks.
+  - category: Use predefined category → Spawns Sisyphus-Junior with category config
+    Available categories:
+  ${categoryList}
+  - subagent_type: Use specific agent directly (explore, librarian, oracle, metis, momus)
+  - run_in_background: true=async (returns task_id), false=sync (waits). Default: false. Use background=true ONLY for parallel exploration with 5+ independent queries.
+  - session_id: Existing Task session to continue (from previous task output). Continues agent with FULL CONTEXT PRESERVED - saves tokens, maintains continuity.
+  - command: The command that triggered this task (optional, for slash command tracking).
+  
+  **WHEN TO USE session_id:**
+  - Task failed/incomplete → session_id with "fix: [specific issue]"
+  - Need follow-up on previous result → session_id with additional question
+  - Multi-turn conversation with same agent → always session_id instead of new task
+  
+  Prompts MUST be in English.`
 
   return tool({
     description,