Skip to content

Products

Crucible is a platform of focused tools sharing a common infrastructure. Each product maps to a stage in the metalworking process.

Product Type Purpose Status
Forge CLI AI environment context compiler ✅ Shipped (v0.6.24)
Ember Collector AI assistant observability 🔬 Designing
Smelt Library Shared plugin engine 📋 Designed
Anvil SDK Plugin development SDK 📋 Designed
Lathe MCP Server Forge commands as MCP tools 📋 Designed
Foundry Web + API Plugin registry/marketplace 📋 Designed
Cast CLI Frontend/embed scaffolding 📋 Designed

The first Crucible product. Today it syncs shared AI assistant configurations from a central source to every AI tool’s expected location. Next: a context compiler that understands skill dependencies, tree-shakes unused content, and optimizes output per agent.

Current (v0.6):

  • 12 commands — init, sync, status, health, config, source management, scheduler, telemetry, uninstall
  • 5 assistants — Claude Code, Cursor, Windsurf, GitHub Copilot, Cline
  • 2 source types — git, local (+ Smithery/agentskills.io coming)
  • 3 layouts — canonical, pre-structured, hybrid (auto-detected)
  • 687 tests, 6,369 LOC production code

Coming (v0.7 — Context Compiler):

  • Skill frontmatter schema with dependencies, tags, context, and priority
  • Dependency graph with topological ordering and cycle detection
  • Project context detection — auto-detect language/framework from marker files
  • Tree-shaking — remove skills irrelevant to the project context
  • Token budgeting — enforce per-agent context window limits
  • Per-agent emitters — optimized output format per assistant
  • Manifest generation — structured build output for Ember telemetry

Forge documentation →

AI Resources Lifecycle Platform — a convention-over-configuration template for building, testing, and delivering AI agent resources.

  • 4 skills, 3 rules, 1 workflow — ready to use or fork
  • 4 meta-skills — orchestrator pattern for designing, developing, and testing resources
  • 4 test specimens — real project snapshots for testing

Starter documentation →

Cross-tool AI assistant observability. Version-aware telemetry for AI coding environments.

Key insight: Forge manages what configs agents get. Ember measures how those configs are used. Together they form a closed feedback loop.

Visibility Gap — without Ember

Forge + Ember Closed Loop

Ember grows progressively in tiers:

Tier Data Source What It Answers
0 (exists) Forge operational events “Is Forge working?”
1 (v0.1) Forge compiler manifest “Which skills, versions, agents per project?”
1.5 (v0.1) OTLP metric export “Same data, in Grafana/DataDog”
2 (v0.2+) Langfuse/Helicone import “Which skill costs how much?”
3 (v0.3+) Agent OTEL + file watchers “Was this skill actually activated?”

Ember Tiered Architecture

Learn more about Ember →

Shared plugin engine library. Will be extracted from Forge when a second consumer (e.g., Cast) exists. No point extracting from one consumer.

SDK for third-party plugin development. Provides types, validation, and testing utilities for building Forge/Cast plugins.

MCP server that exposes Forge commands as MCP tools. Enables AI assistants to run Forge operations directly.

Plugin registry and marketplace. Discover, share, and install community-built skills, rules, and workflows.

Frontend/embed scaffolding CLI. Uses @crucibledx/smelt for plugin infrastructure. Focused on micro-frontend patterns.