aether-colony 3.1.1 → 3.1.3

package/.opencode/commands/ant/ant:verify-castes.md

@@ -1,157 +0,0 @@
----
-name: ant:verify-castes
-description: "Verify model routing is working for all castes"
----
-
-You are the **Queen**. Verify that model routing configuration is correct and test that workers can self-report their assigned models.
-
-## Step 1: Check Proxy Health
-
-Run using the Bash tool: `node bin/cli.js caste-models list`
-
-Capture the proxy status line. If proxy is unhealthy:
-- Display warning about proxy not running
-- Show instructions to start LiteLLM proxy
-- Continue with verification anyway (will show limited functionality)
-
-## Step 2: Verify Each Caste Assignment
-
-For each caste in [prime, builder, watcher, oracle, scout, chaos, architect, archaeologist, colonizer, route_setter]:
-
-1. Get assigned model using the Bash tool:
-   ```
-   node -e "const mp = require('./bin/lib/model-profiles'); const p = mp.loadModelProfiles('.'); console.log(mp.getEffectiveModel(p, 'CASTE').model)"
-   ```
-   Replace CASTE with the actual caste name.
-
-2. Verify model is not "default" (should be specific model from profiles)
-3. Log result with checkmark or X
-
-Display results in a table:
-```
-Caste Verification:
-─────────────────────────────────────────
-✓ prime: glm-5 (z_ai)
-✓ builder: kimi-k2.5 (kimi)
-✓ watcher: kimi-k2.5 (kimi)
-...
-```
-
-## Step 3: Test Worker Self-Reporting (Recommended)
-
-Spawn a test worker from each caste to verify they correctly self-report their model assignment:
-
-### 3.1 Spawn Test Workers
-
-For each test caste in [builder, watcher, oracle]:
-
-1. Get the model assignment:
-   ```bash
-   node -e "const mp = require('./bin/lib/model-profiles'); const p = mp.loadModelProfiles('.'); console.log(mp.getEffectiveModel(p, 'CASTE').model)"
-   ```
-
-2. Spawn test worker via Task tool with `subagent_type="general"`:
-
-**Test Worker Prompt:**
-```
-You are a test worker for model routing verification.
-
---- MODEL CONTEXT ---
-Assigned model: {model} (from caste: {caste})
-Expected: You should process this task using the model assigned above
-
---- YOUR TASK ---
-Simply echo back the model context in the required JSON format.
-
---- OUTPUT ---
-Return JSON:
-{
-  "test_passed": true,
-  "model_context": {
-    "assigned": "{model}",
-    "caste": "{caste}",
-    "source": "caste-default"
-  },
-  "verification": "Model context received and will be echoed back"
-}
-```
-
-3. Collect the worker's JSON response
-
-### 3.2 Display Self-Reporting Results
-
-After all test workers return, display results:
-
-```
-Model Self-Reporting Test:
-─────────────────────────────────────────
-✓ builder: Reported kimi-k2.5 ✓ (matches assignment)
-✓ watcher: Reported kimi-k2.5 ✓ (matches assignment)
-✓ oracle: Reported minimax-2.5 ✓ (matches assignment)
-```
-
-If a worker reports a different model than assigned:
-```
-⚠️ oracle: Reported kimi-k2.5 ✗ (expected: minimax-2.5)
-   → Model routing may not be working for this caste
-```
-
-**Note:** Workers self-report based on the model context in their prompt. This verifies:
-1. Queen correctly assigns models based on caste
-2. Workers receive and can echo the assignment
-3. The routing chain is intact from configuration → spawn → execution
-
-## Step 4: Summary Report
-
-Display final verification report:
-```
-═══════════════════════════════════════════
-Model Routing Verification Complete
-═══════════════════════════════════════════
-Proxy Health: ✓ Healthy (or ✗ Not running)
-Configuration: X/10 castes have model assignments
-Self-Reporting: X/3 test workers echoed correctly
-
-Status: {operational | needs attention}
-```
-
-### If All Checks Pass:
-```
-✓ Model routing is working correctly
-
-All castes have model assignments and workers correctly
-self-report their assigned models. The routing chain:
-  model-profiles.yaml → Queen spawn → Worker context
-is functioning properly.
-```
-
-### If Issues Found:
-```
-⚠️ Issues Detected:
-- Proxy: {status}
-- Missing assignments: {castes without models}
-- Self-report failures: {workers that didn't echo correctly}
-
-Recommendations:
-1. {specific fix for each issue}
-```
-
-## How Model Routing Actually Works
-
-**Current Implementation (Self-Reporting):**
-1. Caste-to-model mappings defined in `.aether/model-profiles.yaml`
-2. Queen reads assignment before spawning worker
-3. Queen includes model context in worker prompt
-4. Worker echoes model_context in JSON output
-5. Queen logs/display shows actual vs expected
-
-**Note on Environment Variables:**
-While the system documents `ANTHROPIC_MODEL` environment variables for
-LiteLLM proxy routing, Claude Code's Task tool doesn't support explicit
-environment variable passing. The self-reporting approach works within
-this constraint by using prompt context instead.
-
-For actual model routing through LiteLLM:
-- Set ANTHROPIC_MODEL in your shell before starting Claude Code
-- Or use the proxy's default model routing
-- Check LiteLLM proxy logs to verify which models are being called