claude-code-slides

Workflows — From Plan Mode to Structured Delegation

Related slides: Slide 15 — Think Before You Build, Slide 16 — Superpowers, Slide 17 — Putting It All Together

Default Plan Mode

Built into Claude Code, no plugins required. Covers ~70% of daily tasks.

Shift+Tab → read-only exploration → generate plan → implement

Pros: Zero setup, fast, editable in your editor.

Cons: Plan lives only in conversation, no spec review, no checkpoints, no parallelism, context bloat as plan + implementation share one window.

Verdict: Great for small-medium tasks. For 5+ files or team review, you need more structure.

Superpowers — Structured Delegation

Superpowers by Jesse Vincent adds rigor to every phase. TDD enforced, sub-agent isolation, spec review before implementation.

claude plugin install --marketplace claude-plugins-official superpowers

The Flow

flowchart TD
    A["/brainstorming"] --> B["Design Spec"]
    B -->|review loop| B
    B --> C["/writing-plans"]
    C --> D["Phases"]
    D --> E["Sub-agents execute\n(TDD per phase)"]
    E --> F["/requesting-code-review"]
    F --> G["/finishing-a-development-branch"]

Key Skills (slash commands)

Skill What it does When to use
/brainstorming Explore requirements, design alternatives, produce reviewed spec Starting any non-trivial feature
/writing-plans Decompose spec into bite-sized tasks with exact file paths and code After spec is approved
/subagent-driven-development Dispatch sub-agents per task, fresh context each, TDD enforced Independent tasks that can parallelize
/executing-plans Sequential execution with checkpoints and review Tasks with dependencies
/test-driven-development Red → Green → Refactor cycle, code before tests gets removed Every implementation task
/requesting-code-review Two-stage review: spec compliance + code quality After implementation
/finishing-a-development-branch Verify tests, present merge/PR/discard options, cleanup Work is done

What Makes It Different

Addresses all plan mode cons: persisted specs, review loops, fresh context per task, parallel execution. See slide 16 for the visual.

Other Structured Workflows

GSD (Get Shit Done)

Fights “context rot” with 5 atomic phases: Discuss → Plan → Execute (parallel waves) → Verify → Ship. Main session stays at 30-40% capacity. Has a quick mode (/gsd:quick) for ad-hoc tasks.

gstack

Virtual engineering team with 15 specialized roles. Pipeline: Think → Plan → Build → Review → Test → Ship → Reflect. Includes browser QA, multi-AI validation, and supports 10-15 concurrent sprints.

Research Workflow: NotebookLM + Obsidian

Before building, you often need to research. Claude Code’s context limit (~200k) can’t hold dozens of sources — but NotebookLM can.

SAP Disclaimer: NotebookLM is not classified as a “productive” tool at SAP — do not upload SAP-confidential information. Use it for public research only.

Why NotebookLM: RAG-based, answers grounded in your actual documents, holds dozens of sources, minimal hallucination. What Claude Code can’t do with context limits, NotebookLM handles natively.

The flow:

Collect sources (NotebookLM) → Query for insights → Take notes (Obsidian) → Feed into Claude Code → Build

# Research
> /nlm Create notebook "Auth Research", add auth0 docs, JWT intro, OAuth2 spec
> /nlm What are tradeoffs between JWT and session-based auth?

# Capture
> /obsidian Create note "Auth Design Decisions" with key findings

# Build
> Read docs/research/auth-decisions.md and plan the auth module

This works for learning, researching, and brainstorming — not just coding.

Tool Link
NotebookLM notebooklm.google.com
nlm CLI github.com/jacob-bd/notebooklm-mcp-cli
Obsidian Skills github.com/kepano/obsidian-skills

Further Reading