πŸ€– AI-Native Development Template

January 8, 2026 Β· View on GitHub

License AI-Powered

A standardized, agentic workflow for modern software engineering. This template configures Claude Code with specialized agents and slash commands to streamline the entire software development lifecycleβ€”from architectural decision-making to high-performance coding.


πŸš€ The "Feature-First" Workflow

We enforce a rigorous yet efficient development loop that prioritizes thinking before coding.

graph LR
    A[Start Feature] --> B[Investigate]
    B --> C{Viable?}
    C -- No --> D[Reject]
    C -- Yes --> E[Draft ADR]
    E --> F[Implement]
    F --> G[Verify & Accept]

1. Exploration 🧭

Start by creating a dedicated feature space. This encourages isolated experimentation. Often I find it helpful and beneficial to do multiple investigations and designs before formalizing a plan. The low level details will stay in the investigation documents while the higher level functional requirements and archtectural elements will be formalized in the ADR.

  • /fn-feature {name}: Scaffolds a new feature folder with a readiness checklist.
  • /fn-investigation {feature} {topic}: Creates a structured document to explore specific technical approaches.

2. Decision Making βš–οΈ

Once investigations yield a clear path, formalize it using Architecture Decision Records (ADRs).

  • /fn-adr {feature}: Synthesizes successful investigations into a Proposed ADR.
  • /fn-reject {feature} {topic}: Formally rejects an investigation path with rationale (crucial for future context).

3. Implementation & Iteration πŸ› οΈ

This is where the actual coding happens.

  • /fn-task {feature} {adr}: The primary engine for implementation. It reads the ADR, breaks it down into sub-tasks, and iterates until the requirements are met.
  • Delegate: Use the specialized agents (see below) through the task command to implement the ADR requirements.

4. Finalization βœ…

  • /fn-review: Technical PR review focused on logic, bugs, and style. It filters false positives and auto-fixes critical items.
  • /wa-review: High-level architectural audit based on the Well-Architected Framework.
  • /fn-document: Triggers the @documentation-agent to update project documentation.
  • /fn-accept {feature} {adr}: Moves the implemented ADR to the permanent record and marks the feature as complete.

🧠 Specialized Agents

Your team of expert sub-agents, each optimized for specific lifecycle phases.

AgentRoleBest ForModel
@adr-analyzerπŸ“ ArchitectParsing ADRs, creating task lists, and verifying implementation compliance.Sonnet
@coding-agentπŸ”¨ EngineerStandard feature implementation, refactoring, and multi-file changes.Sonnet
@complex-coding-agent🧠 Sr. EngineerHigh-complexity architecture, race conditions, and system design.Opus
@fast-coding-agent⚑ Junior DevSingle-file edits, quick fixes, test updates. Fast and cheap.Haiku
@well-architected-agentπŸ›‘οΈ AuditorReviews code against Reliability, Security, Cost, Ops, and Performance pillars.Sonnet
@documentation-agentπŸ“ Tech WriterUpdating docs, fixing typos, and ensuring documentation builds.Sonnet

⚑ Slash Commands

Quick shortcuts to drive the workflow.

Architecture & Feature Management

CommandDescription
/fn-featureInit: Create a new feature folder structure.
/fn-investigationExplore: Add a new investigation topic to a feature.
/fn-rejectDecide: Mark an investigation as rejected (with reasons).
/fn-adrPropose: Generate an ADR from viable investigations.
/fn-taskImplement: Execute and iterate on an ADR's requirements.
/fn-reviewPR Review: Technical review of logic, bugs, and style fixes.
/fn-documentDocs: Update documentation to reflect recent changes.
/fn-acceptFinalize: Promote an ADR to accepted status and move to docs/adr.

Well-Architected Reviews

CommandFocus Area
/wa-reviewAudit: Comprehensive architectural review across 5 pillars.
/wa-securitySecurity: Deep dive into vulnerabilities and auth patterns.
/wa-reliabilityReliability: Check resilience, error handling, and recovery.
/wa-performancePerformance: Analyze scalability, latency, and resource usage.

πŸ› οΈ Usage

Starting a New Feature

> /fn-feature bulk-data-export
> /fn-investigation bulk-data-export stream-serialization
> # ... research ...
> /fn-adr bulk-data-export

Running Reviews

# Architectural Audit
> /wa-review

# Technical PR Review
> /fn-review

Intelligent Coding

Don't just ask to "fix code." Delegate to the right expert, this is often fast and saves tokens:

# Complex refactor or bug fixes (Opus)
> @complex-coding-agent Refactor the connection pooling logic to prevent deadlocks.

# Most coding work (Sonnet)
> @fast-coding-agent Add a null check to the User parser.

# Simple fix (Haiku)
> @fast-coding-agent Add a null check to the User parser.

πŸ“‚ Project Structure

.
β”œβ”€β”€ .claude/
β”‚   β”œβ”€β”€ agents/         # Agent definitions
β”‚   └── commands/       # Slash command documentation
β”œβ”€β”€ docs/
β”‚   β”œβ”€β”€ adr/            # Accepted Architecture Decision Records
β”‚   └── features/       # Active feature development folders
└── src/                # Source code

Also see