Vibe Ecosystem Coding Runtime
A profile-based toolkit for focused Claude Code sessions and bounded Codex work.
- Source repo
- vibeeval/vibecosystem
- Stars
- ★ 553
- Last updated
- 9d ago
- License
- MIT
- Primary language
- C#
- FA score
- 76/100 · Good
At a glance
- How it runs
- Works with
- Portable with changesClaude CodeCodex (Partial support)
- Cost
- Free, no paid service needed
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A Claude Code developer who wants a selected set of tools for investigation, implementation, review, and verification.
- Not a fit if
- Teams needing a turnkey setup across many platforms
- Teams that expect every feature to be enabled by default
- Maintenance
- Commits in the last 30 dayslast commit 2026-11-11latest release v3.5.1 (2026-11-11)+0.6 ★/day over 7 days
- Source review
- 76/100 · Good
What does this agent do, and when should you use it?
vibecosystem packages agents, skills, and hooks for software work, installed through a CLI and selected by profile. The default core profile installs a chosen set for investigation, implementation, review, and verification; quality, context, memory, orchestration, and dashboard are opt-in, while full installs the complete catalog. Claude Code hooks respond to host lifecycle events and can inspect tool input, return a permission decision, or provide bounded context; agent and skill files are instructions consumed by the host model. The repository also has installers for Codex and Cursor, though the Codex luna_worker is an optional bounded worker. The runtime is primarily installed locally, while some optional integrations contact configured services and may store code excerpts, prompts, tool output, or learned rules under ~/.claude/.
Users install the vibeco CLI with install.sh or npx vibecosystem init, then select a profile to install its agents, skills, rules, and hooks. vibeco effective-config shows the effective configuration, vibeco doctor inspects installation state, and vibeco stats reports statistics; commands such as vibeco profile quality switch profiles. When Claude Code fires a registered hook, it can read tool input, return a permission decision, or provide bounded context; the host model decides whether to follow a suggested agent or workflow. Optional context, memory, and orchestration components provide code-context retrieval, persistent recall and learning workflows, and explicit multi-agent workflows. Installing the full catalog does not launch an agent swarm.
- A Claude Code developer who wants a selected set of tools for investigation, implementation, review, and verification.
- A team that wants focused edit, compiler, or security checks can add the
qualityprofile tocore. - A developer working in a large codebase can enable
contextand use a compatible TLDR daemon for code-context retrieval. - A Claude user who needs persistent recall or learning workflows can enable
memoryand configure its dependencies as needed. - A developer setting up selected skills or the optional
luna_workerin Codex can use the Codex installer and worker contract.
Key setup facts
Pulled from the install and usage notes below. The project README has the current values.
- Config files
~/.local/bin- Runtime
- Node.js 18+
How do you install or deploy this agent?
Source installation requires Node.js 18+ and Bash. Clone the repository, review the core install plan with a dry run, then install it. The published npm version can also be initialized with npx. Codex uses a separate installer, and its worker is optional.
How do you use this agent?
Restart Claude Code after changing hook registrations. Inspect the effective configuration and installation state, then switch profiles to match the task. These commands use the documented CLI interface:
vibeco effective-config
vibeco doctor
vibeco stats
vibeco profile quality
vibeco profile coreIf the shell cannot find vibeco, add ~/.local/bin to PATH; the source CLI can also be run directly. Profile changes reconcile installer-owned files and hook registrations. Modified or unmanaged files require review.
What are this agent's strengths and limitations?
- The default
coreprofile installs a selected set for investigation, implementation, review, and verification; the complete catalog is a separate choice. - Profiles separate quality, context, memory, orchestration, and dashboard capabilities, so components and hook registrations can be selected by task.
- Claude Code hook context has character limits: 4,000 per event and 12,000 per session for lean profiles; the README explicitly says these are not total model-input or cost guarantees.
- Separate setup paths are documented for Claude Code, Codex, and Cursor, with CLI inspection commands such as
effective-configanddoctor.
- Requires Node.js 18+ and Bash; profile changes reconcile files and hooks, and modified or unmanaged files need review.
- Some optional features need additional runtimes or services, including a TLDR daemon,
better-sqlite3, Python/uvscripts, the dashboardwspackage, or configured endpoints. - Codex is documented for selected skills and an optional bounded
luna_worker; Claude hook behavior should not be assumed to be native Codex support. - Hooks and agent files provide instructions and suggestions to the host; they do not guarantee that a suggested agent or workflow will run.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Vibe Ecosystem Coding Runtime This agent | 76 · Good | CLIFree | ★ 553 | 9d ago | C# | Claude Code |
| MCO | 75 · Some gaps | CLIFree + model costs | ★ 558 | 14d ago | Python | Codex · Claude Code |
| Concord MCP | 75 · Good | — | ★ 1.4k | 1d ago | TypeScript | Codex · Claude Code |
| Sno Station | 74 · Some gaps | CLIFreemium | ★ 1k | today | TypeScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
SECURITY.md and the README state that hooks run with user privileges, local data may remain under ~/.claude/, optional integrations contact configured services, and the credential filter has limits. The installer documents dry runs, ownership boundaries, and backups before replacement; GitHub writes and Braintrust extraction require explicit environment variables. Deductions: the repository includes optional hooks, services, and persistent local data, while ordinary installation still writes to user directories. The supplied dependency-security evidence does not show a complete dependency audit or handling of known vulnerabilities.
The README, CLI, and validation scripts broadly corroborate core/full component counts and installation flows; the scripts describe isolated-HOME installer checks. Deductions: the README presents Claude and Codex adapters while the repository describes itself primarily as a Claude runtime, and the sample omits implementation details. The CLI doctor checks path existence, which is only a basic health signal. Some errors and skipped operations are documented, but the evidence does not establish clear recovery guidance for every runtime failure.
Documentation offers core, quality, context, memory, orchestration, dashboard, and full profiles, describes Codex, Cursor, and compatible-host scope, and clarifies that full does not automatically start a swarm and hooks do not guarantee workflow execution. Deductions: activation guidance is partly represented by a small sample of skill files, and Node/Bash, Claude-directory, and optional-tool requirements constrain environment fit.
The README organizes setup, profiles, architecture, catalog, adapters, integrations, and development, with links to upgrade, security, verification, and architecture material. Installation steps and dry-run examples are concrete, and the MIT license text is present. Deductions: package and CLI versions agree in the sample, but references to a changelog and upgrade guide alone do not establish a sustained release cadence. A security contact and reporting guidance are present, but fixes depend on maintainer availability and publisher identity is unverified.
Profile-based installation, on-demand skills, context budgets, and a bounded Codex worker provide usable choices for software tasks; the CLI exposes effective-config, doctor, and stats commands. Deductions: utility depends on host models, tasks, optional services, and configuration. The README disclaims a fixed token-saving claim, and the benchmark skill's framework does not itself demonstrate product outcomes.
The README points to repository files and commands for profiles, budgets, integrations, and limitations. It distinguishes character limits from total model tokens, estimates from provider-reported usage, and product diagrams from recordings. The supplied installer validation scripts specify concrete isolated-install checks. Deductions: only a sample of files is available, so not every claim can be cross-checked; static review cannot establish that scripts ran or that their results are correct.
- Before installation, review the profile, write targets, and existing user files. The installer modifies host directories, and replacement backups may retain sensitive content.
- Before enabling memory, dashboard, compiler assistance, MCP, search, or provider integrations, review local storage, credentials, and external data recipients.
- The credential hook is not a sandbox; use host permissions, operating-system isolation, and narrowly scoped credentials for sensitive work.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
Changed - Docs: the README, localized READMEs and runtime overview now show the full catalog separately from the default core. - README marks 3.5.1 as the current npm release. npm: [email protected] (latest). See [CHANGELOG.md](https://github.com/vibeeval/vibecosystem/blob/v3.5.1/CHANGELOG.md)View on GitHub ↗ -
Release notes excerpt
Changed - Claude CI review is opt-in through the claude-review label and cancels stale runs. Ordinary validation does not call an LLM. - Added an explicit dashboard profile for local agent/tool activity monitoring; it is also included in full. - Twenty core skill entrypoints now use bounded workflows with available agents. Entrypoint-byte comparisons and their limits are recorded in the verification guide. …View on GitHub ↗ -
v3.3.0 v3.3.0 - Security, Hook Health, Marketplace
Release notes excerpt
What's New Security Fix - Credential-deny symlink bypass (CRITICAL): Symlinks to sensitive files now correctly blocked via readlinkSync + realpathSync resolution Hook Health Ledger - New hook-health.ts shared module - every instrumented hook reports success/failure/duration - 5 critical hooks instrumented: credential-deny, intent-classifier, model-router, skill-compounder, skill-curator - vibeco doctor --runtime shows 24h execution stats with broken hook detection CI Quality Gates - TypeScript type check (tsc --noEmit) now runs in CI …View on GitHub ↗
FAQ
Does installing full automatically start multiple agents?
full installs the complete catalog and hook registration manifest; it does not launch an agent swarm. Execution depends on the host model and workflow.How many tokens or how much money will it save?
Are memory and dashboard ready to use out of the box?
better-sqlite3 installation, and other recall workflows may need Python/uv and a database. The dashboard uses the local ws package, which must be installed as described in its setup guide.Does it access the network or send data off the machine?
~/.claude/ may contain code excerpts, prompts, tool output, and learned rules. Review enabled profiles, credentials, and endpoints.Does it work the same way in Codex and Claude Code?
luna_worker.Is Vibe Ecosystem Coding Runtime free?
Is Vibe Ecosystem Coding Runtime open source?
Is Vibe Ecosystem Coding Runtime safe to use?
Is Vibe Ecosystem Coding Runtime still maintained?
Who makes Vibe Ecosystem Coding Runtime?
Thanks — we use this to improve the page.