Development workflow

Development Workflow Methodologies

A collection of proven development workflows covering the full lifecycle: from exploration (spikes) through implementation (TDD), verification (code review), cleanup (simplify), and debugging (systematic debugging).

Workflow Sequence

Spike ──→ TDD ──→ Code Review ──→ Simplify ──→ Deploy
  │                                          ↑
  └──→ Bug Found? ──→ Systematic Debugging ──┘

Each phase has an independent skill reference with exact commands, iron laws, and anti-pattern tables:

1. Spike — Throwaway Experiments

When to use: Validating feasibility before committing to a real build.

• references/spike.md for the full workflow

2. Test-Driven Development (TDD)

When to use: ANY code change — features, bug fixes, refactoring.

• RED: Write failing test first, watch it fail
• GREEN: Write minimal code to pass
• REFACTOR: Clean up while keeping tests green
• references/test-driven-development.md for the full TDD cycle

3. Pre-Commit Code Review

When to use: Before every commit after 2+ file edits.

• Static security scan (secrets, injections, unsafe patterns)
• Baseline-aware quality gates (tests, linting)
• Independent reviewer subagent (fail-closed JSON verdict)
• Auto-fix loop (max 2 cycles)
• references/github-code-review.md for the full pipeline

4. Simplify Code — Parallel Cleanup

When to use: When the user says "simplify" or "clean up my changes".

• Three parallel reviewers: Reuse, Quality, Efficiency
• Aggregates findings, applies fixes, verifies with tests
• references/simplify-code.md for the full process

5. Systematic Debugging

When to use: ANY technical bug or test failure.

• 4 phases: Root Cause Investigation → Pattern Analysis → Hypothesis Testing → Implementation
• The Rule of Three: stop after 3 failed fixes, question the architecture
• references/systematic-debugging.md for the full methodology

Common Patterns Across All Workflows

The Iron Law

Each methodology has a non-negotiable first principle:

• TDD: "NO PRODUCTION CODE WITHOUT A FAILING TEST FIRST"
• Debugging: "NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST"
• Code Review: "No agent should verify its own work"

Rationalizations vs Reality

All five skills include a table of common excuses and their counters.

Red Flags

Each skill lists patterns that indicate you're skipping the process.

When NOT to Use Each

• Spike: Skip when the answer is already known or the work is production path
• TDD: Skip only for throwaway prototypes, generated code, or config files (ask the user first)
• Code Review: Skip for docs-only or config-only changes
• Simplify: Skip unless the user explicitly asks (it's token-heavy)
• Debugging: Never skip — even simple bugs have root causes

Hermes Tool Integration

All workflows use the same Hermes tool primitives:

terminal("git diff")              # Get changes
terminal("pytest ...")            # Run tests
terminal("ruff check .")          # Lint
terminal("git log --oneline")     # Check history
search_files("error_string")      # Find errors
read_file("src/problematic.py")   # Read code