ECC 2.2.0 is worth a staged trial if your team already uses Claude Code or Codex and wants a packaged workflow of agents, skills, hooks, memory, and security scanning. It is not a good starting point without an established agent workflow. Before production, install from official channels only, choose one install method per harness, and review hook and rule behavior yourself. This is a source and documentation review as of 31 August 2026, not a runtime test.

Key facts at a glance

What ECC 2.2.0 adds

The maintainers describe ECC as an agent harness performance optimization system: skills, instincts, memory, security, and research-first development for Claude Code, Codex, OpenCode, Cursor, and other coding agents. Their stated workflow is plan, test, implement, review, verify, remember, improve. Version 2.2.0, released 28 August 2026, ships 68 agents, 286 skills, and 94 command shims, plus rules, hooks, memory, continuous learning, and installation surfaces, according to the release notes.

• One guided installer. npx --yes [email protected] install --guided checks every selected harness before writing files and applies the setup as one operation, covering Claude Code, Codex, and Kimi Code.
• Native Antigravity 2.0 support. ECC installs to .agents/{rules,workflows,skills,agents} and migrates legacy ownership, per the release notes.
• Experimental Nasiko CLI bridge. ECC can inspect, install, and remove one pinned Nasiko v0.1.0 CLI build. Downloads are content-addressed and SHA-256 verified, status is read-only, and uninstall protects modified or unqualified binaries. The maintainers state ECC does not connect agents, start a server, enable telemetry, share data, or operate Nasiko’s control plane.
• Stricter release pipeline. The exact npm archive must pass lifecycle tests on Linux, macOS, and Windows before latest moves, and v2.1.0 remains the immediate rollback target.
• Cleaner lifecycle. Repeated selective installs keep an ownership ledger for reliable uninstall, OpenCode now uses ~/.config/opencode, and migration paths preserve edited or untracked files.

These are maintainer claims from the release notes and README. This review verified them by reading repository and release documentation, not by installing or running ECC, so treat runtime behavior as unverified until you reproduce it in your own environment.

Who should consider it

ECC 2.2.0 fits teams that already run a coding agent workflow and want to stop re-prompting the same plan, test, review, and verify process.

• Developers on Claude Code, which the README calls the best-supported harness, who want packaged planning, TDD, review, security, and verification skills.
• Codex users who want the native plugin path (codex plugin marketplace add affaan-m/ECC followed by codex plugin add ecc@ecc), with the caveat that Codex hooks require an explicit trust decision and do not use Claude’s four hook profiles.
• Small engineering teams that want memory, continuous learning, and AgentShield scanning without hand-building their own skill packs.
• Teams planning to run agents against custom or self-hosted model endpoints: ECC works through each harness’s normal configuration, so the prerequisite is a working harness CLI pointed at your gateway.

Who should skip it

• Teams with no established agent workflow. ECC modifies dotfiles, can install hooks, and adds a maintenance surface; its value depends on an existing harness.
• Anyone who needs guarantees rather than claims. This is a documentation review, and the README itself says ECC works best with Claude Code and provides capability-limited adapters elsewhere, so check the support matrix before assuming parity.
• Security-sensitive teams without review capacity. The hooks runtime only materializes with an explicit decision, but once enabled, hooks can run commands; AgentShield scanning is a maintainer claim, not an independent audit.
• Teams that cannot absorb a weekly release cadence or single-maintainer risk. The README states a single maintainer ships weekly across harnesses, and the 120 open issues at snapshot are a reason to test, not a verdict.

What to check before production

The safe path is: verify the source, confirm versions, pick one install method, review hooks and rules, and test in a disposable project before touching a shared environment.

• Source control first. Install only from the official GitHub repository, the ecc-universal npm package, the GitHub App, the ecc@ecc plugin slug, or ecc.tools. The README warns that third-party re-uploads are not maintained or reviewed and may contain malware; if npm reports a version or cache error, confirm the registry version before retrying.
• Version requirements. The universal package needs Node.js 18 or newer; the Claude plugin path also needs Git and Claude Code 2.1 or newer on PATH.
• One install method per harness. Guided setup, native plugin, and manual install are alternatives, not layers. The maintainers say stacking methods duplicates skills, commands, hooks, or configuration; if you already layered installs, use the reset or uninstall flow rather than reinstalling.
• Review hooks and rules. Any profile or module that materializes the hook runtime requires an explicit decision; use a minimal or no-hooks profile if you do not want runtime hooks. Claude Code plugins cannot distribute rules, so copy only the rule packs you actually use.
• Scope and permissions. Decide project-local vs global configuration, and audit what hooks, MCP configuration, and permissions will be allowed. The release notes state that provider-owned model and authentication settings stay under the provider’s control.
• Private repositories. The OSS repository is MIT, but ECC Pro is the hosted GitHub App for private repos, listed at $19/seat/month in the README.
• Upgrade and rollback. v2.1.0 is the stated immediate rollback target; review the changelog between v2.1.0 and v2.2.0 before upgrading existing installs, and treat the Nasiko and Itô bridges as experimental until validated.

Where custom work and maintenance are needed

Even with the guided installer, ECC is configuration, not a turnkey product. The places teams typically need help are:

• Building or trimming rules and skills for your actual stack; the README suggests starting with rules/common plus one language or framework pack.
• Wiring custom or self-hosted model endpoints and any model-name remapping in the harness configuration.
• Cleaning up legacy manual or sync-based installs and resolving conflicting plugin scopes before a clean reinstall. If you are inheriting an agent setup with layered installs or stale config, that cleanup is an upgrade-and-migration task rather than a fresh install.
• Keeping up with weekly releases and harness version drift; the README notes, for example, that Cursor-native loading behavior can vary by build.
• Enforcing an MCP connector policy and deciding which connectors are allowed, since AgentShield scans MCP configuration, permissions, secrets, and agent files.

Alternatives and evaluation questions

ECC is one way to standardize agent behavior. The alternatives are native harness plugin systems, your own minimal rules and skills, or a narrower internal pack built for one workflow; the right choice depends on how much configuration you want to own.

• Which harnesses and versions does your team actually run, and does each support ECC natively or only via a capability-limited adapter?
• Who will own installs, upgrades, and rollbacks, and can they keep up with a weekly release cadence?
• Do you need the hooks runtime, memory, and continuous learning, or would a minimal rules-and-skills install cover the workflow?
• Will agents talk to official providers, a custom gateway, or self-hosted models, and is that path already working outside ECC?
• Are you on private repositories where ECC Pro pricing applies, and does that change the cost case?

Verdict

On the evidence, ECC 2.2.0 is a reasonable candidate for teams that already run Claude Code or Codex, want a packaged plan-test-review-verify workflow, and can test before rolling out. The conditions are non-negotiable: install from official channels, use one install path per harness, review what hooks and rules will do, and validate in a sandboxed repository first. The experimental bridges (Nasiko, Itô) are not production features yet. Because this is a documentation review, your own runtime test is the deciding step before you standardize on it.

Sources

• ECC repository on GitHub
• ECC v2.2.0 release notes
• ecc-universal on npm

What I can do next

If you are weighing ECC for a team, I can help with the evaluation itself: auditing your current Claude Code, Codex, or Cursor setup, planning a single-path install or a clean migration away from older manual configs, and building the rules, skills, and hook configuration your stack actually needs. To do that properly I would need access to your environment or a representative repository, the harness versions you run, and a disposable test project; I would not promise a productivity result until we see how the setup behaves on your own workloads. If the decision is bigger than one plugin, our software development services cover agent integration and internal tooling, and you can contact XP812 with your harness versions and current setup.