ECC: The Project Formerly Called Everything Claude Code

ECC

ECC is a community-maintained, open-source agent-harness project. It began as Everything Claude Code and now presents a wider system of instructions, rules, skills, commands, agents, hooks, memory, security tooling, installers, and orchestration for Claude Code, Codex, OpenCode, Cursor, Gemini-oriented environments, and other agent tools.

ECC is not an Anthropic product, an OpenAI product, or an official cross-harness standard. Its source is available under the MIT license at affaan-m/ECC.

This profile covers ECC 2.0.0 at commit 6a9f075, committed July 26, 2026. The main branch changed after that point, so current installation and release claims require a fresh review.

Product at a glance

The project calls the 2.0 line an agent-harness operating system. That name reflects its breadth, not a formal operating-system category.

From Everything Claude Code to ECC

Venture Step episode 111A records the earlier name. Dalton Anderson explores Everything Claude Code as a repository of patterns that could improve an agent's access to project context and reusable guidance.

The project later moved to the shorter ECC name and expanded its stated audience. Its cross-harness architecture distinguishes native, instruction-backed, adapter-backed, and reference-only capabilities rather than claiming that every tool behaves like Claude Code.

That distinction is important. A Claude Code hook can run on a supported native event. A similar Codex package may express the intention as instructions without providing the same event enforcement. Shared content does not guarantee shared execution semantics.

What version 2.0.0 contains

The pinned checkout identified itself as package ecc-universal version 2.0.0. The project's 2.0.0 release material describes 261 skills, a control-pane foundation, worktree lifecycle, and orchestration components.

Local source-provided validators counted 67 agent files, 94 command files, 122 rule files, 281 validated skill directories, 21 hook matchers, 34 install modules, 81 install components, and seven profiles. The difference between the release's skill count and the validated directory count reflects measurement definitions and repository structure. Counts should always be dated and labeled.

The repository is much larger than its former name suggests. It contains configuration for several harnesses, executable Node scripts, installer paths, optional MCP definitions, memory features, evaluation tools, worktree services, and an experimental Rust-based second-generation runtime area.

flowchart TD
    A["ECC source"] --> B["Rules and instructions"]
    A --> C["Skills, commands, and agents"]
    A --> D["Hooks and scripts"]
    A --> E["MCP and runtime config"]
    A --> F["Install and harness adapters"]
    A --> G["Memory and orchestration"]
    B --> H["Selected coding harness"]
    C --> H
    D --> H
    E --> H
    F --> H
    G --> H

Each arrow is a trust path. A user should know which ones are active in a given installation.

How ECC differs from official harness documentation

Official product documentation defines what a harness currently supports. ECC packages opinions and implementations on top of those products.

Anthropic's Claude Code extension guide is the authority for current Claude behavior across CLAUDE.md, rules, skills, subagents, hooks, MCP, and plugins. OpenAI's AGENTS.md guide and skills guide control current Codex behavior.

ECC can provide a useful implementation example, but it cannot make one harness support another harness's native events. Its own compatibility documentation generally acknowledges that limit.

Installation and adoption

ECC offers profiles and explicit modules rather than only one monolithic path. The reviewed installer source and tests include dry-run behavior and support for multiple targets.

In a focused local run, the install-apply test suite passed 32 cases. Those tests included minimal profiles, dry runs, managed install state, preservation of existing settings in named scenarios, conflict reporting, and Claude, Cursor, Antigravity, JoyCode, and Qwen paths.

That evidence supports the behavior of the tested commit in the review environment. It does not guarantee that an installer will preserve every custom repository. Run the dry plan first, begin from a committed state, inspect the resulting diff, and keep a recovery path.

Selective adoption is the safest use. A team can study one rule structure, validation script, review agent, or workflow and adapt it to a known need. Copying the complete repository creates responsibility for every active hook, script, connection, and update.

Hooks and automatic behavior

ECC's hook documentation describes pre-tool, post-tool, session-start, pre-compaction, stop, and session-end behavior. The pinned graph includes quality checks, context persistence, summaries, pattern evaluation, cost markers, and notifications.

Hooks can be valuable because they make recurring checks automatic. They also execute code at moments a user may not be actively watching. Review the matcher, command, environment, timeout, exit behavior, output, and supported target before enabling one.

Anthropic's current hooks guide explains that event behavior differs by hook type and that post-tool hooks cannot reverse an action. Users should compare ECC's configuration with the current official semantics.

MCP, memory, and connected systems

MCP configurations can give an agent access to external tools or data. Memory systems can preserve material across sessions. Both features change the privacy and authority boundary.

Review whether a connector is enabled by default or opt-in, which package or command launches it, which endpoint it reaches, what credentials it reads, what data it sends, and what actions it exposes. Remove servers that do not support a real workflow.

Review memory storage in the same way. Determine the location, scope, retention, source labels, access controls, and deletion path. A local Markdown format can be easier to inspect than an opaque store, but local storage is not automatically safe.

Security position

Google's analysis of files that coding agents trust treats repository instructions, runtime settings, extensions, and executable configuration as part of the developer attack surface.

That framework applies to ECC and every comparable project. It does not imply malicious intent. It means that natural-language files, hooks, MCP definitions, installers, and extensions deserve the same source review and least-privilege design as conventional software.

The pinned ECC validators passed their named structural checks, and the harness audit reported 31 checks with none failing. Those results are useful project evidence. They are not an independent security certification.

Best fit

ECC fits experienced agent-tool users who want a broad pattern library or a starting point for an internal harness. It is most useful when the adopter already has project-specific tests, permission boundaries, version-control discipline, and someone who owns the configuration.

It is less suitable as a first configuration for a small repository with no established instructions. Official documentation and one compact project file may solve that case with less complexity.

Organizations that cannot review third-party scripts and connections should not enable a broad community pack. They can still study the public architecture and implement a smaller internal equivalent.

Why ECC appears on Venture Step

Episode 111A uses the earlier Everything Claude Code repository to explore a change in AI-assisted development. The key idea is that repeatable work depends on context architecture and verification, not one clever prompt.

The episode is a historical source for Dalton's experience. This profile keeps the public entity current and separates that experience from the product's present name and scope.

Read [[ECC Review - What to Borrow and What to Review First|ECC Review: What to Borrow and What to Review First]] for the dated evaluation. [[How to Structure Repository Instructions for Coding Agents]] turns the architectural lesson into a tool-neutral guide.

Verification and disclosure

This profile was checked on July 27, 2026 against ECC 2.0.0 at commit 6a9f075cd97c139a5f7e84e1e3f2c9ab095adf64, current GitHub metadata, repository source and documentation, focused local validators and installer tests, official Anthropic and OpenAI documentation, Google's security analysis, and the E111A transcript.

Repository contents, release status, ownership, commercial offerings, compatibility, dependencies, popularity, and installation behavior require periodic refresh. AI assisted with research organization and drafting; evidence boundaries and final editorial decisions remain Dalton Anderson's.