Integration test failure: microsoft-foundry – quota model capacity tracking [Assertion mismatch]

## Summary

**Run:** [Integration Tests - all #130](https://github.com/microsoft/GitHub-Copilot-for-Azure/actions/runs/23833768263)
**Test:** `microsoft-foundry_quota - Integration Tests › Monitor Quota Across Deployments › explains capacity by model tracking`
**Test File:** `tests/microsoft-foundry/quota/integration.test.ts:178`
**Date:** 2026-04-01
**Skill:** `microsoft-foundry`

## Diagnosis

**Root Cause Category:** Assertion mismatch

The test asserts that the assistant message includes **both** the keyword `"model"` AND the keyword `"capacity"` when responding to a quota-by-model query:

````typescript
const hasModelTracking = doesAssistantMessageIncludeKeyword(agentMetadata, "model")
&& doesAssistantMessageIncludeKeyword(agentMetadata, "capacity");
expect(hasModelTracking).toBe(true); // line 178 — received: false
```

The agent correctly invoked the `microsoft-foundry` skill and returned a quota table with model breakdown (e.g., `OpenAI.Standard.gpt-4o: 10 TPM used / 59 TPM limit`) — but used the terms **"limit"**, **"available"**, and **"headroom"** instead of **"capacity"**. The word `"capacity"` did not appear in the assistant message, causing the `&&` assertion to fail.

## Prompt Context

```
Show me quota allocation by model in Azure AI Foundry
````

**Skill routing:** `microsoft-foundry` ✅ (correct)
**Agent behavior:** Queried `az rest` Azure Management API, recovered from a JMESPath syntax error using `jq`, and returned a structured quota table grouped by model.
**What was missing:** The response described quota allocation using "limit/used/available" terminology but never used the word "capacity" in the assistant message text.

## Skill Report Context

From the SKILL-REPORT for run #130:

| Test | Type | Runs | Pass Rate | Avg Confidence |
|------|------|------|-----------|----------------|
| `capacity_by_model_tracking` | End-to-End | 1 | 0% (0/1) | — |

The skill report notes the agent auto-recovered from a JMESPath error and presented comprehensive data, but the assertion on response vocabulary failed.

## Environment

| Property | Value |
|----------|-------|
| Run ID | [23833768263](https://github.com/microsoft/GitHub-Copilot-for-Azure/actions/runs/23833768263) |
| Run # | 130 |
| Branch | main |
| Commit | e948e29 |
| Test duration | 61.8s |
| Azure subscription | GithubCopilotForAzure-Testing |

## Recommended Fix

Either:
1. **Update the test** to use `"limit"` or `"quota"` as the second keyword (matching the actual response vocabulary from the skill), or
2. **Update the skill guidance** to ensure the response explicitly uses the word "capacity" when summarizing quota allocation by model (e.g., in the quota.md reference file for microsoft-foundry).

> Generated by [Analyze Test Run](https://github.com/microsoft/GitHub-Copilot-for-Azure/actions/runs/23835729644) · [◷](https://github.com/search?q=repo%3Amicrosoft%2FGitHub-Copilot-for-Azure+is%3Aissue+%22gh-aw-workflow-call-id%3A+microsoft%2FGitHub-Copilot-for-Azure%2Fanalyze-test-run%22&type=issues)