Dan Robbins
Windows · C# / .NET · T-SQL · story planning · multi-repo orchestration
Most of my Claude time splits between two things: planning stories for the Eagles team and delivering PBIs across a 91-repo ecosystem that spans PowerBuilder, .NET APIs, Azure Functions, T-SQL databases, and Service Bus consumers. The planning side is where the leverage is — a thorough interrogatory loop before implementation means the coding agents downstream get self-contained task specs with zero ambiguity. By the time a story hits the sprint board, every “OR” condition is resolved, every stored procedure is named, and every column type is specified. The implementation is mostly mechanical after that.
Codebases
Where I spend my time, roughly in order:
- C# / .NET — APIs (ASP.NET Core, MediatR, Dapper), Azure Functions (V4, net8.0), CSO consumers, and transmission processors. This is the bulk of the new work.
- T-SQL / SQL Server — Stored procedures, schema migrations, and data validation across TranAMS, CorpRes, and AuctionAccess. I know the AMS SP naming conventions (
spams_s_,spams_m_,spams_i_,spams_u_,spams_d_,spams_v_,spams_x_) cold. - PowerBuilder — Legacy AMS frontend (ASG-AMS-Suite, VLC-AMS-Suite). Still actively maintained, still in production. Modernization happens around it, not instead of it.
- Azure DevOps Pipelines — YAML pipeline definitions for build, test, and deploy across the repo fleet.
MCPs
The ones I keep connected:
- SQL Server (
carvana-sql) — Query test/UAT database instances directly from Claude. Lets me verify schema, check column types, and validate SP behavior without leaving the terminal. This is the one I use the most. - IDE (
ide) — VS Code integration for diagnostics and code execution. Useful for running quick checks without switching windows. - Azure DevOps — Work item creation, linking, and management. The story planning workflow creates PBIs and tasks through this.
- Context7 — Current library docs so Claude isn’t guessing based on stale training data.
- Google Drive — Reading specs, meeting notes, and reference materials that live in shared drives.
Plugins
Two installed, both from the org marketplaces:
ws-atlas(Wholesale marketplace) — The Wholesale Atlas knowledge base. Maps the entire Eagles ecosystem: 91 repos across 13 categories, AMS stored procedure inventories, CSO message schemas, trigger chains, App Insights components. When I need to find which repo owns a feature or trace a dependency chain, this is faster than grepping across 56 local clones.essentials(Carvana marketplace) — Development workflow tooling. Code review, implementation planning, debugging, PR descriptions. The/essentials-analyzeand/essentials-create-planskills get the most use.
Skills I use the most
Project-local (Eagles-story-planning repo)
/discover-feature— Kicks off a full feature discovery. Searches codebases, queries DB schemas, identifies patterns, and produces a shape-of-work before anyone writes a line of code./discover-story— Takes a single story from the shape-of-work and produces the complete spec: behavioral acceptance criteria, per-repo task breakdowns with file paths, and a validation plan./implement-team— Orchestrates multi-repo implementation against task specs. Coordinates branch creation, coding agents, and PR workflows across repositories./process-feedback— Pulls session reflections from merged PRs and analyzes them against current rules, proposing harness improvements./ado-sync— Syncs story planning artifacts with Azure DevOps work items./yaml-pipeline— Generates Azure DevOps YAML pipeline definitions.
Plugin-provided
/deliver-pbi— Full PBI delivery lifecycle with phased execution: discovery, agent spawning, PR creation, monitoring, simplification, documentation, and audit. Ties everything together for end-to-end feature delivery./ws-atlas— Query the Wholesale Atlas for repo ownership, dependency maps, SP inventories, and system documentation./chrysler-migration— Chrysler transmission migration automation./test-http-trigger— End-to-end HTTP trigger testing./run-ado-pipeline— Create, find, and monitor ADO pipeline runs.
Workflow
Sessions usually start from the story planning repo, even when the eventual change is a one-line fix somewhere else.
- Discover. Gather reference material (briefs, ADO features, PRDs, meeting notes) into a working directory. Run
/discover-featureto build a knowledge base and produce a shape-of-work. - Interrogate. Iterate with Claude on gaps, ambiguities, and open questions. Five stages: open discovery, context integration, scope/constraints, technical details, validation. Don’t move past this until every “OR” is resolved.
- Spec. Run
/discover-storyfor each story. Produces behavioral acceptance criteria, per-repo task breakdowns with exact file paths and implementation specs, and a validation plan with goalposts. - Deliver. Developer picks up the story, runs
/deliver-pbior/implement-team. Coding agents execute against the task specs. PRs get created, reviewed, and merged. - Close the loop.
/process-feedbackscans merged PRs for session reflections and proposes harness improvements — rule updates, agent prompt revisions, template tweaks.
Settings
- Model:
claude-opus-4-6. Planning quality matters more than token economy. - Permissions:
skipDangerousModePermissionPrompt: true. Same reasoning as Patrick — the permission prompts become noise when you’re running 50+ queries per session. - Marketplaces: Both the Wholesale and Carvana marketplaces are configured.
- Platform: Windows 11, bash shell, VS Code.
Why this works for me
- Planning is the multiplier. A detailed story spec eliminates back-and-forth during implementation. The coding agents don’t need to ask questions because the spec already answered them.
- The SQL MCP is the secret weapon. Half of story planning is verifying what already exists in the database — column types, SP signatures, trigger chains. Being able to query the actual schema from inside the conversation instead of alt-tabbing to SSMS saves enormous context-switching cost.
- 91 repos is manageable with the right index. The ws-atlas plugin means I don’t have to remember which repo owns what. I query the atlas, get the answer, and move on.
- Close-the-loop feedback compounds. Every sprint, the harness gets slightly better at producing specs that don’t need revision. The investment in
/process-feedbackpays off across the whole team, not just my sessions.