Every Monday, the team meets to review results from weekend compete runs -- CLI, Claude Code, and Codex CLI evaluated across Sonnet 4.6, Opus 4.7, and GPT 5.4. Today, this meeting is largely passive reporting: scores go up, scores go down, the team discusses possible reasons, and then disperses to work on whatever feels most promising.
What's missing is a closed loop:
Weekend Runs --> Identify Signals --> Select Investments --> Execute --> Validate
A deterministic analysis pipeline that takes raw agent trajectory recordings (ATIF v1.6 format) and produces structured reports showing exactly what the agent did, how much it cost, and what went wrong.
ATIF trajectory.json → OTel spans → normalize → metrics + diagnostics + turns → reports → repo mapping → code fixes / sub-issues
Brainstorm document — April 2026
Agent skills today are agent-generated through user interaction and by service SMEs, producing agent-consumable markdown grounded in Microsoft Learn documentation. This works well for individual skills, but leads to inconsistencies in content coverage across services — some skills have deep SDK references and troubleshooting tables while others have minimal coverage. The Microsoft Learn MCP tools (microsoft_docs_search, microsoft_docs_fetch, microsoft_code_sample_search) already exist and return rich, structured content — but are inconsistently used in skill authoring today, with only a few skills leveraging them while most rely on manual knowledge capture.
A practical guide to adding telemetry to your VS Code extension -- covering setup, data captured, and PII handling.
@vscode/extension-telemetry answers "what are my users doing and what's breaking?" It's built for product analytics and error monitoring -- the 80% case for extension authors.
azure-prepare plans infrastructure by analyzing source code and mapping it to one of 5 host types (containerapp, appservice, function, staticwebapp, aks). It has no logic to plan the remaining 43 resource types in this skill — VNets, Firewalls, VPN Gateways, VMs, Service Bus, Key Vault, etc. It cannot look at a workload description and decide "you need a hub-spoke VNet with NSGs and a Firewall."
This skill fills that gap: infrastructure-first planning for platform engineers, where the input is a workload description (not code) and the output covers all 48 resource types. Renaming to azure-enterprise-infra-planner makes this distinction clear.
Progressive Disclosure for MicrosoftDocs/Agent-Skills via azure-skills — single azure-docs skill design with 3-tier loading
Date: March 12, 2026 Problem: Agent-Skills has 193 service skills, each 100-850 lines. Loading them all is impossible. Loading even one full skill (e.g., Azure Functions at 366 lines / ~135 links) wastes tokens when the user only needs one category. Solution: Restructure Agent-Skills into a three-tier progressive disclosure model with a single router SKILL.md, per-service reference files, and on-demand Microsoft Learn fetching.
Date: March 12, 2026 Scope: How
azure-prepareuses recipes, runtimes, and service references through progressive disclosure Walkthrough prompt: "Create a Node.js Express API with Cosmos DB that I can deploy to Azure Container Apps"
Date: March 11, 2026 Context: Beyond diagnostics/troubleshooting, where else do azure-skills and Agent-Skills overlap? What content from Agent-Skills should azure-skills reference?
There are 6 major overlap areas between azure-skills (24 operational skills) and MicrosoftDocs/Agent-Skills (193 service documentation skills):
azure-skills + MicrosoftDocs/Agent-Skills Convergence: Diagnostics & Troubleshooting Strategy
Date: March 11, 2026 Context: Should
azure-skillsreferenceMicrosoftDocs/Agent-Skillscontent? Should diagnostics be split into diagnostics + troubleshooting?