Skip to content

Latest commit

 

History

History
230 lines (182 loc) · 12.3 KB

File metadata and controls

230 lines (182 loc) · 12.3 KB

Enchanted Plugins Roadmap

Vision: Build the algorithm-driven operating system for AI-assisted development. 20 plugins connected through an MCP client, each backed by a named formal algorithm.

Architecture Phases

Phase 1 (NOW)          Phase 2               Phase 3               Phase 4
5 plugins              MCP Client POC        10 plugins            Production MCP
Individual installs    Unified interface     Full coverage         Real-time dashboard
                                                                   Developer adoption

Foundation Infrastructure

Before the first plugin ships, the ecosystem needs one shared piece: the repo template every sibling is cloned from.

Repo Role Status
enchanter-ai/schematic Canonical repo template. Ships the 8-section CLAUDE.md, 10 shared/vis/conduct/*.md modules, docs/architecture/ auto-generation pipeline, plugins/example-subplugin/ skeleton, renderer toolchain. Never installed — cloned. Shipped

The template is the contract. When it drifts, all downstream siblings drift — so changes to schematic are treated as brand-standard changes, not per-plugin improvements.


Phase 1: Core 5 Plugins (Foundation)

The first 5 plugins answer the 5 fundamental questions of AI-assisted development.

# Plugin Question Algorithms Version Status
1 Wixie What did I say? (prompt quality) Gauss Convergence, Boolean SAT, Game Theory, Cross-Domain Adaptation v3.0.0 Shipped — 7 plugins
2 Emu What did I spend? (token health) Markov Drift, Shannon Compression, Linear Runway, Atomic Serialization v2.0.0 Shipped — 3 sub-plugins + 1 meta
3 Crow What just happened? (change comprehension) Bayesian Trust, Semantic Diff, Information-Gain, Session Continuity v1.0.0 Shipped — 5 plugins
4 Hydra Is it safe? (security) Aho-Corasick, Shannon Entropy, Config Poisoning, Phantom Dependency, EMA Posture Decay v1.0.0 Shipped — 6 plugins
5 Pech What did it cost? (spend tracking) Exponential Smoothing, Budget Forecasting Not started

Milestone: 5 plugins shipped

  • Each plugin is a standalone Claude Code marketplace
  • Each follows @enchanter-ai brand standard
  • Each has named algorithms, managed agents, self-learning
  • Users install individually: /plugin marketplace add enchanter-ai/<name>

Phase 2: MCP Client POC (Unification)

Build enchanted-mcp — a Model Context Protocol client that connects all 5 plugins into a single orchestration layer.

What the MCP Client Does

enchanted-mcp
├── Connects to all installed enchanter-ai via MCP
├── Unified dashboard: prompts + tokens + changes + security + costs
├── Cross-plugin intelligence:
│   ├── Wixie detects bad prompt → Emu shows token waste from it
│   ├── Crow flags risky change → Hydra scans it for vulnerabilities
│   ├── Pech shows cost spike → traces to which plugin/session caused it
│   └── All learnings shared across plugins (Gauss Accumulation network)
└── Single install: `npx enchanted-mcp` or Docker container

Architecture

┌──────────────────────────────────────────────┐
│              enchanted-mcp (client)           │
│                                              │
│  ┌─────────┐ ┌─────────┐ ┌─────────┐       │
│  │  Wixie   │ │  Emu  │ │  Crow  │  ...  │
│  │  (MCP)  │ │  (MCP)  │ │  (MCP)  │       │
│  └────┬────┘ └────┬────┘ └────┬────┘       │
│       │           │           │              │
│  ┌────▼───────────▼───────────▼────┐        │
│  │     Cross-Plugin Intelligence    │        │
│  │     Shared learnings.json        │        │
│  │     Unified event bus            │        │
│  └──────────────────────────────────┘        │
│                                              │
│  ┌──────────────────────────────────┐        │
│  │     Dashboard (localhost:3000)    │        │
│  │     Real-time session overview    │        │
│  └──────────────────────────────────┘        │
└──────────────────────────────────────────────┘

Milestone: POC MCP client

  • Connects to Wixie + Emu + Crow + Hydra + Pech
  • Shared event bus for cross-plugin signals
  • Basic web dashboard showing unified session view
  • Cross-plugin learnings (Gauss Accumulation network)

Phase 3: 10 Plugins (Full Coverage)

Add 5 more plugins covering code quality, testing, DevOps, documentation, and API design.

# Plugin Question Algorithm Category
6 Athena Is this code good? AST Diff + Weighted Decision Trees Code review
7 Crucible Do the tests catch bugs? Genetic Mutation Testing Testing/QA
8 Assembler Can this deploy? Critical Path DAG Optimization DevOps/CI
9 Scribe Is the docs up to date? TF-IDF Extractive Summarization Documentation
10 Schema Is the API contract valid? Semantic Version Diffing API design

Milestone: 10 plugins + enhanced MCP

  • All 10 plugins connected to enchanted-mcp
  • Dashboard shows full development lifecycle
  • Cross-plugin intelligence covers: prompt → code → test → security → deploy → docs
  • Plugin-to-plugin event triggers (Crow flags change → Athena auto-reviews → Crucible tests)

Phase 4: Production MCP (Developer Adoption)

15 More Plugins

# Plugin Algorithm Category
11 Beacon Isolation Forest Anomaly Detection Observability
12 Nexus Topological Sort + Dependency DAG Multi-repo
13 Comply SPDX License Graph Resolver Compliance
14 Prism WCAG Rule Engine + axe-core Accessibility
15 Tempo Statistical Flame Graph Sampling Performance
16 Rosetta Levenshtein Fuzzy Deduplication i18n
17 Onboard Spaced Repetition (SM-2) Learning
18 Synapse CRDT Knowledge Merging Collaboration
19 Vault Three-Way Merge Diffing Database
20 Relay Event Sourcing + Saga Pattern Webhooks
21 Sylph Jaccard-Cosine Boundary Segmentation + Myers-Diff Conventional Classifier Git workflow (shipped early — v0.0.1, 9 plugins)

Production MCP Features

  • Real-time web dashboard with WebSocket updates
  • Team mode: shared learnings across developers
  • Cost alerts and budget enforcement
  • Plugin marketplace within the MCP (install/remove from dashboard)
  • API for external integrations (Slack, Linear, Jira)
  • Telemetry and analytics (opt-in)
  • Plugin SDK for third-party developers

Milestone: 21 plugins + production MCP

  • Full developer operating system
  • Every stage of AI-assisted development covered
  • Algorithm-driven, agent-managed, self-learning at every layer
  • Active developer community
  • Third-party plugin ecosystem

Timeline

Phase Milestone Plugins Target
1 Foundation 5 (Wixie, Emu, Crow, Hydra, Pech) — 4/5 shipped (Pech not started) Q2 2026
2 MCP POC 5 + MCP client Q3 2026
3 Full Coverage 10 + enhanced MCP Q4 2026
4 Production 21 + production MCP — Sylph (#21) shipped early Q1 2027

Naming Convention

Every plugin is named after a game entity that metaphorically describes its function.

Plugin Entity Game Why
Wixie Enchantment Orbs Minecraft XP orbs that power the enchantment table — enchanting prompts
Emu Emu Mob Minecraft Flying creature that collects items and brings them to you — collecting tokens
Crow Crow Alex's Mobs Silent watcher that observes every change, scores trust with a Bayesian eye, and flags what looks wrong — watching changes
Hydra Hydra Twilight Forest Many-headed boss that regenerates and never truly dies — every secret scanning, vuln, and supply-chain check it survives makes it stronger
Pech Pech Thaumcraft Small scholarly creature obsessed with counting and cataloguing every resource it touches — cost accounting
Athena Athena Hades Goddess of wisdom who judges your combat quality and grants boons for excellence — code review
Crucible Crucible Terraria Endgame crafting station forged in hellfire — tests things to destruction — mutation testing
Assembler Assembling Machine Factorio Takes parts in, produces artifacts out, chains into automated pipelines — CI/CD building
Sylph Sylph Ars Nouveau Airy nature spirit that weaves arcane threads into coherent flows — weaving branches, commits, and PRs into one history

Brand Standard (All Plugins)

Every @enchanter-ai product must:

  1. Name every engine after a formal algorithm
  2. Delegate background work to managed agents (Opus/Sonnet/Haiku)
  3. Persist learning across sessions (Gauss Accumulation)
  4. Report honest numbers — never inflate claims
  5. Use atomic operations and handle race conditions
  6. Maintain zero external dependencies (bash + jq for hooks, Python stdlib for scripts)
  7. Include tests, dark-themed PDF reports, and comprehensive documentation
  8. Follow the Emu-style plugin marketplace structure

Algorithm Registry

Every named algorithm across the ecosystem:

ID Name Product Engine
F1 Gauss Convergence Wixie Standard deviation minimization
F2 Boolean SAT Overlay Wixie Hybrid SAT + continuous optimization
F3 Cross-Domain Adaptation Wixie Constraint-preserving model translation
F4 Game-Theoretic Security Wixie Zero-sum adversarial robustness
F5 Static-Dynamic Verification Wixie Structure + behavior dual testing
F6 Gauss Accumulation Wixie Cross-session knowledge persistence
A1 Markov Drift Detection Emu Hidden state transition recognition
A2 Linear Runway Forecasting Emu Token consumption prediction
A3 Shannon Compression Emu Information-theoretic output reduction
A4 Atomic State Serialization Emu Bounded checkpoint persistence
A5 Content-Addressable Dedup Emu Hash-based read deduplication
V1 Semantic Diff Compression Crow Multi-file change clustering
V2 Bayesian Trust Scoring Crow Prior-posterior change risk assessment
V3 Information-Gain Decision Crow Review prioritization by uncertainty reduction
V4 Session Continuity Graph Crow Decision-causal relationship persistence
V5 Adversarial Self-Review Crow Specific concern generation for risky changes
V6 Gauss Learning (Crow) Crow Developer preference accumulation
S1 Aho-Corasick Pattern Hydra Multi-pattern secret scanning
S2 Shannon Entropy Analysis Hydra High-entropy string detection
L1 Exponential Smoothing Pech Cost forecasting
L2 Budget Boundary Detection Pech Spend threshold alerting
W1 Myers-Diff Conventional Classifier Sylph Diff-to-Conventional-Commits classification
W2 Jaccard-Cosine Boundary Segmentation Sylph Task-boundary clustering from edit-event stream (defining engine)
W3 Workflow-Pattern Classifier Sylph Repo-signal → branching-model inference
W4 Path-History Reviewer Routing Sylph Blame-graph reviewer suggestion
W5 Gauss Learning (Sylph) Sylph Developer workflow-preference accumulation

This is a living document. Update as plugins ship and algorithms evolve. Sylph W1–W5 are seed names from prompts/sylph-architecture/ — final names emerge from the architecture prompt's output.