Workflow Mastery

Workflow Mastery for .NET

Core Principles

1. Parallel over sequential — Run 3-5 Claude sessions simultaneously using git worktrees. Build a feature in one, fix a bug in another, run tests in a third. The single biggest productivity unlock.
2. Plan then execute — For any non-trivial task, start in plan mode, iterate until the plan is bulletproof, then switch to auto-accept. A good plan means Claude 1-shots the implementation.
3. Verification closes the loop — Give Claude a way to prove its work: dotnet build, dotnet test, get_diagnostics via MCP. This single practice 2-3x the quality of the output.
4. Context is a budget, not a dumping ground — The context window fills fast: a typical .cs file is 500-2000 tokens, and 50 file reads can burn a large share of the budget. Spend tokens like sprint capacity — deliberately.
5. Automate the repetitive — If you do it more than once a day, make it a hook, a slash command, or a subagent. Pre-allow safe permissions. Eliminate friction.
6. Compound your knowledge — Every correction becomes a rule in MEMORY.md (see instinct-system skill). Every PR review adds a learning. Over time, Claude's mistake rate drops because your project's knowledge base grows.

Patterns

Parallel Sessions with Git Worktrees

The biggest productivity multiplier. Each worktree gets its own Claude session, its own files, zero conflicts.

# Create worktrees for parallel work
git worktree add ../my-project-feature origin/main
git worktree add ../my-project-bugfix origin/main
git worktree add ../my-project-tests origin/main

# Start Claude in each (separate terminal tabs)
cd ../my-project-feature && claude
cd ../my-project-bugfix && claude
cd ../my-project-tests && claude

Practical .NET workflow:

Tips:

• Name your terminal tabs by task so you never lose track
• Use shell aliases (alias zf='cd ../my-project-feature') for one-keystroke switching
• Enable terminal notifications so you know when a session needs input

Auto-Format Hook for .NET

Catch formatting issues on every file write — eliminates the "CI failed on formatting" loop.

// .claude/settings.json
{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Write|Edit",
        "hooks": [
          {
            "type": "command",
            "command": "dotnet format --include \"$CLAUDE_FILE_PATH\" --no-restore 2>/dev/null || true"
          }
        ]
      }
    ]
  }
}

Why || true: The hook should never block Claude's workflow. If formatting fails (e.g., on a non-C# file), silently continue.

Pre-Allow Safe .NET Permissions

Stop clicking "allow" for every dotnet command. Add these to .claude/settings.json:

{
  "permissions": {
    "allow": [
      "Bash(dotnet build *)",
      "Bash(dotnet test *)",
      "Bash(dotnet run *)",
      "Bash(dotnet ef *)",
      "Bash(dotnet format *)",
      "Bash(dotnet restore *)",
      "Bash(dotnet pack *)",
      "Bash(dotnet tool *)"
    ]
  }
}

Check this into git so the whole team gets frictionless workflows.

Plan Mode Strategy

For any task touching 3+ files or involving architecture decisions:

Step 1: Enter plan mode (Shift+Tab twice)
Step 2: Describe the task with full context
Step 3: Iterate on the plan — challenge assumptions, ask "what about edge cases?"
Step 4: Once the plan is solid, switch to normal mode
Step 5: Claude executes with auto-accept — often 1-shots the implementation

Advanced pattern: Have one Claude write the plan, then spin up a second Claude session to review it as a staff engineer:

"Review this plan as a staff .NET engineer. Challenge every assumption.
What could go wrong? What's missing? What would you do differently?"

When things go sideways: The moment implementation deviates from the plan, STOP. Don't push through. Switch back to plan mode, understand what changed, re-plan, then resume.

Verification Loop for .NET

For the full 7-phase verification pipeline (build, diagnostics, anti-patterns, tests, security, format, diff review) with structured PASS/FAIL reporting, see the verify skill.

Boris's #1 tip: "Give Claude a way to verify its work." The short version: always tell Claude to run dotnet build, dotnet test, get_diagnostics, and dotnet format --verify-no-changes before declaring done. The verify skill has the complete pipeline with short-circuit rules and report templates.

Compounding Knowledge via Corrections

For the full correction capture system — detection, generalization, categorized storage, and periodic audits — see the `instinct-system` skill. The short version: after every correction, capture a generalized rule in MEMORY.md so the same mistake never recurs.

Prompting Techniques for .NET

Challenge Claude's work:

"Grill me on these changes. Would this pass a staff .NET engineer's code review?
Check for: N+1 queries, missing CancellationToken, exposed domain entities,
missing validation, incorrect service lifetimes."

Demand proof:

"Prove this works. Run the tests, show me the output.
Then diff the API response between main and this branch."

After a mediocre fix:

"Knowing everything you know now, scrap this and implement the elegant solution.
No hacks, no workarounds."

For EF Core migrations:

"Generate the migration, then show me the raw SQL it produces.
I want to verify the migration before applying it."

Subagent Patterns for .NET

The kit ships 10 specialist agents — route to them before writing your own: dotnet-architect, code-reviewer, refactor-cleaner, test-engineer, security-auditor, build-error-resolver, ef-core-specialist, api-designer, performance-analyst, devops-engineer. Each carries pre-loaded skills and domain context a generalist session lacks.

Use them:

"Run the code-reviewer agent on my changes before I create the PR."
"Have refactor-cleaner simplify the files I just modified."
"Send the failing CI log to build-error-resolver."

For workflows the kit does not cover, create project-specific subagents in .claude/agents/ — a markdown file with a role, a numbered job list, and a required report format (PASS with summary / FAIL with specifics). Keep one concern per agent so its output stays reviewable.

When to offload vs. stay in main context: see Context Discipline below — subagents are also your context isolation chambers, not just task runners.

Context Discipline

The rules in .claude/rules/agents.md already mandate MCP-first navigation (find_symbol before file reads, get_diagnostics over builds). This section is the strategy layer on top: how to budget, when to offload, and how to recover.

Token Economics

A Roslyn MCP query costs 30-150 tokens; a file read costs 500-2000+. To understand OrderService, four MCP calls (find_symbol → get_public_api → find_references → get_type_hierarchy) cost ~310 tokens; reading the four related files costs ~2900. Then read only the method you'll modify. Reserve full file reads for files you are about to edit.

Subagent Offloading Decision Matrix

OFFLOAD TO A SUBAGENT WHEN:
- Exploring unfamiliar code (> 3 files to read)
- Research requiring docs or multiple files
- Verbose output (test runs, diagnostics, comparisons)
- Any task where the journey is verbose but the answer is concise

STAY IN MAIN CONTEXT WHEN:
- Modifying a file you've already read
- Quick lookups (1-2 MCP queries)
- Work that builds on the ongoing conversation with the user