AI that codes like a senior engineer. Checklists, quality gates, and verification built into every workflow.
Experimental - This plugin is under active development. We are fine-tuning subagent orchestration to ensure reliable skill loading and phase execution. We will add GitHub releases once the plugin stabilizes.
| Command | Purpose | When to Use |
|---|---|---|
/code-foundations:research |
Clarify what you want through facilitated conversation | Exploring ideas |
/code-foundations:plan |
Create implementation-ready plans | Feature planning |
/code-foundations:building |
Execute plans with quality gates | Implementing approved plans |
/code-foundations:debug |
Scientific debugging with task tracking | Bug hunting |
Why this exists: LLMs write code fast. Fast code without engineering discipline creates debt. This plugin loads proven checklists and mental models so Claude applies them automatically.
Three commands work together: Research clarifies intent, Plan creates the plan, Building executes it.
/code-foundations:research "add notification system"
↓
.local/state/code-foundations/research/2026-01-30-notifications.md
↓
/code-foundations:plan .local/state/code-foundations/research/2026-01-30-notifications.md
↓
.local/state/code-foundations/plans/2026-01-30-notifications.md
↓
/code-foundations:building .local/state/code-foundations/plans/2026-01-30-notifications.md
Facilitated conversation to extract and document requirements.
User: "/code-foundations:research add user notifications"
→ Facilitated conversation
→ Progressive narrowing: purpose, actors, context, boundaries, needs, risks
→ Save confirmed requirements to .local/state/code-foundations/research/
Researches your codebase, audits available skills, then asks targeted questions.
User: "/code-foundations:plan add user notifications"
DISCOVER
├─ Search codebase for existing patterns
├─ Audit ALL available skills (from every installed plugin)
│ → "React Native project detected → react-native-foundations:coding"
│ → "Frontend UI work → design-for-ai:a11y-audit"
├─ Ask targeted questions (one at a time)
└─ Produce problem statement
EXPLORE (Medium/Complex)
├─ 2-3 structurally different approaches
└─ Pre-mortem (Complex)
DETAIL → SAVE → CHECK → CONFIRM → HANDOFF
├─ Phase specs with Skills field per phase
├─ Save plan to .local/state/code-foundations/plans/
├─ Subagent reviews plan with fresh eyes
├─ User confirms + corrections
└─ Handoff to /code-foundations:building
Skills loaded: aposd-designing-deep-modules, aposd-reviewing-module-design
Task tracking: Creates progress tasks at startup so you can see where plan is in its flow.
Gated execution with subagents. Each phase has mandatory quality checks.
User: "/code-foundations:building .local/state/code-foundations/plans/2026-01-30-notifications.md"
BRANCH GATE
└─ On main? → STOP. Create feature branch first.
FOR EACH PHASE:
┌────────────────────────────────────────────────────────────┐
│ BUILD Build-agent: discovery + design + TDD │
│ ⛔ Loads pre-gate + implement standards │
├────────────────────────────────────────────────────────────┤
│ REVIEW Post-gate-agent checks quality (Full gate) │
│ ⛔ Cannot commit until reviewer returns PASS │
├────────────────────────────────────────────────────────────┤
│ COMMIT Orchestrator commits after gates pass │
└────────────────────────────────────────────────────────────┘
| Phase | Standards Loaded | What Gets Enforced |
|---|---|---|
| BUILD | references/pre-gate-standards.md + references/implement-standards.md |
Design-before-code, interface depth, control flow, naming, complexity |
| REVIEW | references/post-gate-standards.md |
Correctness, quality, module design, error handling |
| VERIFY | performance-optimization, cc-refactoring-guidance |
Performance regressions, refactoring opportunities, build + tests + lint |
Gate policy is adaptive: Full (BUILD + REVIEW), Standard (BUILD + tests), Minimal (BUILD only). Skills assigned per phase during plan's SAVE step.
The system saves every artifact to .local/state/code-foundations/building/. Per-phase commits enable rollback.
Predict, log, run, resolve. Task list keeps you on track.
/code-foundations:debug login fails 20% of the time
TASK #1: Investigate login failure
├─ PREDICT: "All tokens should be valid"
├─ LOG: Add at validateToken entry
├─ RUN: 2 of 10 fail, tokens valid
└─ RESOLVE: Problem is downstream → narrow
TASK #2: Narrow: validateToken result
├─ PREDICT: "Cache should HIT on second call"
├─ LOG: Add at cache check
├─ RUN: Two MISS within 10ms
└─ RESOLVE: Race condition found → fix
TASK #3: Fix: request deduplication
└─ RESOLVE: Fix applied → verify
TASK #4: Verify: parallel logins succeed
└─ RUN: 100 parallel → 0 failures → Done!
Skill loaded: cc-debugging (scientific debugging method)
The task list prevents rabbit holes, missed verifications, and lost context.
| Situation | Command |
|---|---|
| Vague idea, unclear requirements | /code-foundations:research |
| Need full feature planning | /code-foundations:plan |
| Have approved plan, ready to implement | /code-foundations:building |
| Bug hunting, need structured approach | /code-foundations:debug |
# Add marketplace
/plugin marketplace add ryanthedev/rtd-claude-inn
# Install
/plugin install code-foundations@rtd
# Update
/plugin update code-foundations@rtdBased on Code Complete, 2nd Edition by Steve McConnell and A Philosophy of Software Design by John Ousterhout.
MIT