Everything Claude Code Release Notes: Complete Version History (1.8.0 – 1.9.0)

Everything Claude Code versions 1.8.0 and 1.9.0 introduced manifest-driven installation, support for 10+ programming ecosystems, and autonomous loop operators protected by a 5-layer safety guard.

The Everything Claude Code (ECC) repository at affaan-m/everything-claude-code maintains a detailed CHANGELOG.md that serves as the canonical source for all release notes. While the VERSION file tracks an internal 0.1.0 identifier for tooling purposes, the public release cycle officially began with version 1.8.0, followed rapidly by the ecosystem-expanding 1.9.0 release. Both versions follow the project's strict test-driven, security-first, immutable development model documented in AGENTS.md.

Version 1.9.0 (March 20, 2026) – Ecosystem and Architecture Expansion

Version 1.9.0 represents a major architectural shift toward modular installation and comprehensive language coverage, adding 6 new agents and dozens of specialized skills.

Manifest-Driven Installation and State Management

The release replaced monolithic installation with a manifest-driven pipeline controlled by install-plan.js and install-apply.js in the scripts/ directory. This architecture enables selective component installation, allowing developers to install only specific agents or skills rather than the entire suite. An SQLite-backed state store (state-store.js) provides deterministic tracking of installed components, ensuring repeatable deployments across environments.

Multi-Language Agent Support (10+ Ecosystems)

ECC expanded beyond its initial scope to support Rust, Java, Kotlin, PyTorch, and existing ecosystems through specialized resolver and reviewer agents:

• typescript-reviewer – Dedicated TypeScript and JavaScript code review
• java-build-resolver and java-reviewer – Maven/Gradle error resolution and Spring Boot analysis
• kotlin-reviewer and kotlin-build-resolver – Full Kotlin ecosystem support
• rust-reviewer and rust-build-resolver – Cargo and Rust toolchain integration
• pytorch-build-resolver – CUDA, runtime, and training error handling
• docs-lookup – Context7-powered documentation research assistant

Invoke any agent through the CLI:

claude-agent run rust-build-resolver Cargo.toml

New Slash Commands and Skills

Three new commands entered the core workflow:

• /docs – Triggers the docs-lookup skill for API reference queries (e.g., claude /docs react@18)
• /aside – Opens isolated side conversations without polluting main session context
• /prompt-optimize – Executes the prompt-optimizer skill to refine user inputs using the everything-claude-code self-referential skill

New domain-specific skills include pytorch-patterns, nextjs-turbopack, mcp-server-patterns, data-scraper-agent, and ai-regression-testing, bringing the total catalog to 109 validated skills.

Observer Reliability and the 5-Layer Loop Guard

To prevent runaway autonomous operations, version 1.9.0 implemented memory throttling, sandbox fixes, and a 5-layer loop guard in the observer subsystem. This safety mechanism monitors agent loops across five distinct checkpoints, automatically terminating sessions that exceed resource or iteration thresholds.

Version 1.8.0 (March 4, 2026) – Harness-First Reliability

Version 1.8.0 established the harness-first foundation, focusing on evaluation discipline, hook reliability, and autonomous loop mechanics.

Hook Runtime and Profile-Based Control

The release introduced profile-based hook control, allowing developers to enable or disable specific hooks via environment variables. Key infrastructure changes included fixing SessionStart root resolution and moving session summary persistence to the Stop hook event. New hooks added were quality-gate (enforcing test coverage ≥ 80% and lint-clean status) and cost-tracker (monitoring API spend during long operations).

Activate a specific profile before running commands:

export ECC_HOOK_PROFILE=ci && claude /harness-audit

NanoClaw v2 and Model Routing

The NanoClaw v2 subsystem debuted with advanced model-routing capabilities, skill hot-loading, conversation branching, and search compaction. The /model-route command allows dynamic switching between Claude model variants based on task complexity, while improved metrics export enables detailed performance analysis.

Autonomous Loop Operators

Two specialized agents enabled safe long-running automation:

• loop-operator – Manages continuous execution with built-in safety guards
• harness-optimizer – Tunes configuration for maximum reliability

The /loop-start command initiates monitored loops with configurable intervals:

claude loop-start --name data-pipeline --interval 30s

Additional command additions included /loop-status for health checks and /quality-gate for merge-blocking verification.

Key Source Files Referenced in Release Notes

According to the affaan-m/everything-claude-code source structure, these files define the systems described above:

• CHANGELOG.md – Human-readable release history and version dates
• AGENTS.md – Canonical registry of all 27 agents and core principles
• scripts/install-plan.js and scripts/install-apply.js – Manifest installation logic
• scripts/state-store.js – SQLite state management for components
• agents/ – Directory containing agent definitions (e.g., typescript-reviewer.md)
• skills/ – Directory with 114 skill markdown files
• commands/ – Slash-command implementations (e.g., harness-audit.md)
• hooks/ – Hook definitions driving harness and loop behavior

Summary

• Version 1.9.0 (March 20, 2026) introduced manifest-driven installation via install-plan.js, SQLite state tracking, 10+ language ecosystems including Rust and PyTorch, and the 5-layer loop guard for observer safety.
• Version 1.8.0 (March 4, 2026) delivered the harness-first architecture with profile-based hooks (quality-gate, cost-tracker), NanoClaw v2 model routing, and the loop-operator agent for autonomous execution.
• Both versions maintain strict immutability and security standards, with all changes documented in CHANGELOG.md and agent specifications stored in the agents/ directory.
• Public releases began at 1.8.0 despite the VERSION file showing internal identifier 0.1.0.

Frequently Asked Questions

What is the latest stable version of Everything Claude Code?

Version 1.9.0 (released March 20, 2026) is the current stable release, adding manifest-driven installation and support for Rust, Java, Kotlin, and PyTorch ecosystems. It succeeds version 1.8.0, which established the harness-first reliability foundation.

How do I upgrade from version 1.8.0 to 1.9.0?

Run the selective install pipeline introduced in 1.9.0 using install-plan.js to generate a component manifest, then apply it with install-apply.js. The SQLite-backed state-store.js will migrate your existing configuration deterministically without breaking 1.8.0 harness settings.

What is the 5-layer loop guard in version 1.9.0?

The 5-layer loop guard is an observer safety mechanism that monitors autonomous agent loops across five distinct checkpoints—including memory throttling and sandbox boundaries—to automatically terminate runaway processes before they exhaust resources or enter infinite recursion.

Where are the release notes documented in the source code?

All release notes are recorded in CHANGELOG.md at the repository root, with agent-specific changes detailed in individual files under agents/ and skill updates in skills/. The AGENTS.md file provides the canonical agent registry and version compatibility matrix.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}