⚡ CapInt

Capability-First Orchestration Engine for IDE Agents
Bridging natural language intents to precise skills and structured workflows.

CapInt is an IDE-native, capability-first orchestration engine designed for developer agents. It allows developer tools (like Cursor, Claude Code, Gemini, and Antigravity) to resolve user intent to specific capability plans and workflows without memorizing skill names.

[!TIP] New to CapInt? Read GUNLUK.md (3 steps, no terminal) or the full Turkish guide in docs/kullanim-kilavuzu.md to get started!

🗺️ Architectural Workflow

graph TD
    User([User Prompt / Natural Language]) -->|Fuzzy Intent| IP[intent-parser.js]
    IP -->|Compound Split| T1[Task 1]
    IP -->|Compound Split| T2[Task 2]
    T1 & T2 -->|Fuzzy Match / Levenshtein| CR[capability-router.js]
    CR -->|Read Matrix| SRM[skill-routing-matrix.json]
    SRM -->|Registry Check| REG[registry.json]
    REG -->|Match Skill / Workflow| EP[execution-policy.js]
    EP -->|Context Memory Bridge| CMB[context-memory-bridge.js]
    CMB -->|Output Contract| Final([Execution Intent: Capability + Provider + Plan])
    
    style User fill:#3498db,stroke:#2980b9,stroke-width:2px,color:#fff
    style Final fill:#2ecc71,stroke:#27ae60,stroke-width:2px,color:#fff
    style SRM fill:#f1c40f,stroke:#f39c12,stroke-width:2px,color:#333
    style REG fill:#e67e22,stroke:#d35400,stroke-width:2px,color:#fff

✨ Features at a Glance

| Feature | Description | Status | | :--- | :--- | :---: | | Fuzzy Matching | Typo-tolerant keyword matching using Levenshtein distance (75% threshold). | v0.7.1+ | | Compound Intent Splitting | Automatically splits complex requests into discrete subtasks (e.g. ve, and, then). | v0.7.1+ | | Workflow Engine | Supports linear phase execution, branching, loops, and context bridge. | v0.9.x | | Contextual Learning | Remembers recent routes, applying a routing confidence bias (+2%/+5%). | v0.7.1+ | | Phase Gates & Validation | JSON schema validation on phase outputs and input gates. | v0.9.x | | MCP Server Mode | Exposes CapInt as a standard Model Context Protocol (MCP) server. | v0.9.x | | Skill Market | Remote skill installation, updates, and list from a registry. | v0.7.1+ | | Interactive Mode (REPL) | Interactive shell with tab-completion and history. | v0.7.1+ |

🚀 Quick Start

# 1. Initialize CapInt in your project (copies 105 skills and 17 workflows)
npx @bcelep/capint init

# 2. Test routing for a natural language request
npx @bcelep/capint route "i18n çeviri lang file hub"

# 3. Open UI Wizard (terminal optional for daily work)
npx @bcelep/capint ui --open

[!IMPORTANT] npx @bcelep/capint init creates AGENT.md, AGENTS.md, design.md, a customized .gitignore fragment, the routing matrix, the registry, and clones all skills to skills/ and workflows to workflows/. By default, it automatically runs capint ide sync to configure your IDE rule files.

🛠️ CLI Command Reference

Yönlendirme (Routing)

• capint route "<task>" — Displays the Execution Intent.
• capint route --verbose "<task>" — Displays the full routing table and rules to read.
• capint route --json "<task>" — Output as JSON.
• capint route --list — Lists all registered capability mappings.

Batch Routing:

• capint batch intents.txt — Route multiple intents from a file.
• echo -e "debug\nrefactor" | capint batch - — Route intents from standard input.

Workflow Chaining:

• capint chain list — List capability chains.
• capint chain show <capability> — Show details of a specific chain.
• capint chain run <capability> [text] — Execute a full chain.

Yönetim ve Durum (Operations & Status)

• capint status — Checks matrix and registry health.
• capint doctor — Drift diagnosis and integration check.
• capint audit — Project and contract audit.
• capint consult "<task>" — Recommendation only (read-only, dry-run).

IDE Entegrasyonu (IDE Integration)

• capint ide sync — Re-synchronize IDE configurations (Cursor .mdc, Claude CLAUDE.md, Gemini GEMINI.md, Antigravity .agents/rules).
• capint ide check — Verify the health of IDE rules.

Yaşam Döngüsü ve Yetenekler (Lifecycle & Skills)

• capint skill list — List all skills on disk.
• capint skill disable <name> / capint skill enable <name> — Toggle skills.
• capint skill pin <name> — Pin a skill to prioritize it.
• capint skill graph — Visualize skill dependency graph.
• capint skill refresh — Force refresh the registry based on disk skills.
• capint recover --list | --latest [--apply] — Restore matrix/registry state.
• capint upgrade --apply — Smart matrix merge, registry refresh, and skill/workflow sync.
• capint uninstall --yes — Completely uninstall CapInt from the project.

Yapılandırma ve Önbellek (Config & Cache)

• capint config list / capint config set <key> <value> — Configure settings.
• capint cache stats — View cache hits, misses, and response times.
• capint cache clear — Evict the routing cache.

🏗️ Core Architecture & Contract

CapInt resolves prompts into a structured Execution Intent contract:

{
  "intent": "i18n çeviri lang file hub",
  "capability": "localization-hub",
  "resolution": "skill/localization-hub",
  "memory": "optional",
  "plan": "confirm",
  "verification_required": true
}

Source Structure

• src/lib/intent-parser.js — Fuzzy matching and compound intent splitting.
• src/lib/capability-router.js — Skill collection and provider resolution.
• src/lib/execution-policy.js — Confirm blocks, locale-handling, and thresholds.
• src/lib/context-memory-bridge.js — Memory strategies.
• src/lib/route-engine.js — Main routing and caching engine.

🧑‍💻 Maintainer Notes

Run tests locally before publishing:

npm install
npm test

[!WARNING] Do not run capint init in the repository itself. Use npm run capint -- <cmd> for dogfooding CLI commands locally. See docs/maintainer-dogfood.md for details.