Runx Skill Workflow Runtime

Run agent skills under explicit authority and leave a verifiable record of each execution.

Source repo
runxhq/runx
Stars
★ 104
Last updated
1d ago
License
Apache-2.0
Primary language
Rust

At a glance

How it runs
CLIMCP serverLibrary / SDK
Works with
Universal · cross-platformCodex · Claude Code (Partial support)
Cost
Free, no paid service needed
Setup effort
Medium · a few setup steps
You'll need
GitNode.js 20 or newerShell / CLINetwork accessLocal filesystemMCP Server
Typical use
Engineering teams can route bounded code plans into repository work with github-issue-to-pr and retain receipts for each run.
Not a fit if
  • Teams that require OS-level isolation for local processes
  • Teams that need the runtime to authenticate human approvers
Website
runx.ai ↗ set on GitHub
Maintenance
Commits in the last 30 dayslast commit 2026-11-20latest release cli-v0.11.1 (2026-11-02)
Source review
82/100 · Good

What does this agent do, and when should you use it?

Runx is a Rust runtime for skill workflows, with a CLI, an MCP server, and TypeScript/Node.js and Python clients. Skills are portable packages centered on SKILL.md; packages that need executable steps can declare methods, inputs, outputs, and capabilities in run.yaml. At each call boundary, the runtime checks authority, delivers credentials, supervises execution, and signs an Ed25519 receipt covering execution, permissions, and artifact provenance. Execution, state, and receipt signing can run locally; hosted registry, credential custody, and notary services are optional. It suits teams that need composable skills and reviewable execution records, with the caveat that local process tools retain host OS permissions.

Run runx skill to resolve a skill from a local directory, URL, or registry and execute a method described by its SKILL.md and run.yaml. Workflows compose agent, tool, operation, ask, and skill steps, passing typed inputs and outputs. A run can pause for the host to provide agent answers or human approvals, then continue with runx resume. Connector operations execute against declared request, scope, and effect contracts; credentials are delivered only at the admitted transport boundary. At completion, Runx creates a signed receipt that can be inspected with runx history and checked with runx verify.

  1. Engineering teams can route bounded code plans into repository work with github-issue-to-pr and retain receipts for each run.
  2. Release owners can compose release, documentation, and messaging skills to prepare launch materials and advance them under approval requirements.
  3. Marketing teams can combine lead-enrichment, lead-qualify, and cold-outreach to assemble evidence and prepare one email for review.
  4. Growth teams can combine Search Console, GA4, PostHog, and site evidence into ranked priorities and a follow-up measurement plan.
  5. Platform teams can expose an in-progress skill turn over MCP to an existing agent host, then resume the same run with its response.

Key setup facts

Pulled from the install and usage notes below. The project README has the current values.

Install / run
  • npm install --global @runxhq/cli
  • curl -fsSL https://runx.ai/install | sh
Config files
answers.json
Runtime
Node.js 20
Install notes cover
macOS · Linux

How do you install or deploy this agent?

Install the native CLI with npm, or use the standalone installer on macOS/Linux. The repository's local example also requires Git and Node.js 20 or newer; it does not require a Runx account, model, or provider key. The installer fetches content from runx.ai; using registries, connectors, or hosted services requires the corresponding network access and configuration.

npm install --global @runxhq/cli

Standalone installer for macOS/Linux:

curl -fsSL https://runx.ai/install | sh

How do you use this agent?

Clone the repository and run its included hello-world skill. The command executes the checked-in Node.js script and returns JSON with a receipt ID; then inspect the run and verify the full receipt ID. This example needs no model or provider key.

git clone --depth 1 https://github.com/runxhq/runx.git
cd runx

runx skill ./examples/hello-world --run-id first-run \
  -i message="Hello, Runx." --json

runx history first-run --json
runx verify <receipt-id> --json

To let an existing agent host handle paused turns, serve the run over MCP and resume it with answers to its pending requests:

runx mcp --run <run-id>
runx resume <run-id> answers.json --json

What are this agent's strengths and limitations?

Pros
  • The CLI, MCP server, and language clients share one Rust execution path, making it practical to integrate with an existing agent host.
  • Calls require explicit authority, and credentials are delivered at the operation transport boundary rather than placed in skill instructions or model context.
  • Ed25519 receipts capture permissions, execution outcomes, child-run links, and artifact provenance, and can be verified independently.
  • Local execution and state do not depend on hosted services; registry, credential custody, and notary services are optional.
Limitations
  • Local process tools run with host OS permissions; Runx does not provide portable OS-level isolation.
  • The host is responsible for obtaining human decisions; JSON in an approval field does not authenticate the approver.
  • Real connector work requires account or credential setup, and fixture-backed tests do not establish live provider behavior.
  • The smallest repository example requires Git and Node.js 20 or newer; broader adoption also involves configuring skill packages and authority.

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
Runx Skill Workflow Runtime This agent 82 · Good CLIFree ★ 104 1d ago Rust —
REA Reverse Engineering Assistant 54 · Major gaps CLIFree + model costs ★ 115 1mo ago Rust Claude Code
Flow-Like 74 · Some gaps Desktop appFreemium ★ 981 1d ago Rust —
Agency Orchestrator: One Sentence to Orchestrate an AI Expert Team 71 · Some gaps CLIFree + model costs ★ 2.5k 10d ago TypeScript Codex · Claude Code · OpenAI API · Claude API

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Good
82/ 100 5-point scale 4.1 / 5
Trust 21/29
Reliability 9/14
Adaptability 16/18
Convention 16/18
Effectiveness 12/13
Verifiability 8/8
Why each dimension lost points
Trust21 / 29 · 3.6/5

The material describes authority narrowing across calls, credentials delivered only at boundaries, and receipts that exclude raw secrets. It explains workflow authority, data flow, and external effects in detail. Approval is obtained by the host, and an approval field alone does not prove human identity; process tools run with host OS permissions and are not OS isolated. The packet does not show a complete recovery or revocation mechanism for external effects, and it provides no dependency audit or vulnerability status evidence, so those areas are scored down.

Reliability9 / 14 · 3.2/5

The CLI, client, runtime, receipt model, and examples form a largely coherent product account, with structured failures and offline verification described. The supplied excerpts do not establish dependency availability or runtime error behavior; some workflows depend on hosted services and external CLIs, so full marks are not supported.

Adaptability16 / 18 · 4.4/5

The material covers local skills, workflows, MCP, clients, hosted connectors, and research, release, and analytics scenarios. It states the boundaries of process tools, host isolation, and hosted services, and describes when Runx is not needed. Platform prerequisites are stated, but the sample cannot establish support across all environments, so environment fit is scored 2.

Convention16 / 18 · 4.4/5

The README, getting-started, credentials, security, and runtime contract documents have clear roles. The packet includes prerequisites, install commands, examples, limitations, an Apache-2.0 license, and a rolling security-fix policy. The package version is 0.0.0 while the README shows an npm version badge; the packet lacks a clear changelog and verifiable maintainer identity, reducing naming, versioning, and maintenance scores.

Effectiveness12 / 13 · 4.6/5

Typed inputs and outputs, signed receipts, authority proofs, and concrete workflow examples make outputs usable for follow-on calls and review, with governance and traceability value beyond ordinary prompt orchestration. The runtime, host, and optional hosted components add setup and deployment costs; no cost measurements are supplied, so cost-benefit does not receive full marks.

Verifiability8 / 8 · 5.0/5

Key README claims correspond to the security policy, authority proof, getting-started guide, and supplied client/CLI excerpts. The material distinguishes receipt integrity from provider truth and host authentication, and scopes guarantees across local and hosted paths. This is an incomplete static sample and cannot establish execution outcomes; the score concerns traceability and separation of fact from inference, not runtime correctness.

Risks and how to mitigate them
  • Process tools are not isolated by a Runx OS sandbox; review local commands according to host OS permissions.
  • Approval records do not prove approver identity; the host must authenticate the human. Receipt signatures matter only when the verifier trusts the corresponding signing key.
  • This review uses only a static source sample; it did not execute the project or verify dependencies, actual permission enforcement, or recovery from external effects.
Evidence confidence: Low Reviewed Nov 13, 2026 Reviewed revision 14cf84bcc80d New commits since this review; the score may not cover them
See the full review method →
Maintain this project? Add the FARS badge to your README

The badge shows the current review score and verdict and links back to this page. It refreshes about once an hour after a re-review.

FARS

[![FARS review](https://followagents.com/badge/runx.svg)](https://followagents.com/en/agents/runx)
<a href="https://followagents.com/en/agents/runx"><img src="https://followagents.com/badge/runx.svg" alt="FARS review"></a>

Recent releases

Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.

  1. cli-v0.11.1
    Release notes excerpt
    Runx CLI 0.11.1 Local registry refreshes now respect the existing record's write permissions while retaining atomic publication. This patch also updates transport dependencies and prepares distribution through the established trusted npm publisher. Highlights - A refused registry update leaves the existing record intact. - Updated MCP and HTTP dependencies retain the native execution boundary. Added - Windows coverage for atomic registry publication, retained readers, no-clobber creation and file access. Changed …
    View on GitHub ↗
  2. cli-v0.11.0
    Release notes excerpt
    Runx CLI 0.11.0 Runx 0.11.0 keeps every secret in one place, runx secret, and a connector whose credential's variable is set needs no setup at all. A name without a version runs at the newest version your runx reads, so a package published for a newer runx no longer breaks an older one. What runx adds to a project now lives under .runx, apart from the skills and connectors you write. Connectors can run as a person's signed-in browser account, and the payment skills settle through Stripe and PayPal as well as x402. Highlights …
    View on GitHub ↗
  3. runx-py-v0.1.2
    Release notes excerpt
    Full Changelog: https://github.com/runxhq/runx/compare/cli-v0.10.2...runx-py-v0.1.2
    View on GitHub ↗

FAQ

Does a run require a model or a Runx account?
The repository's hello-world example requires neither a Runx account, a model, nor a provider key. Delegating agent turns with --drive requires a configured model provider, and live connector operations need the relevant accounts or credentials.
Does Runx authenticate human approvers?
No. Runx separates approvals from agent answers and binds each decision to the request digest, but the host obtains the human decision. A JSON value submitted by the host does not establish identity.
What does a receipt prove?
It records skill and execution digests, authority, outcomes, and artifact links, with a content-addressed ID and signature for verification. A valid signature establishes integrity against a trusted signer; it does not mean a run succeeded or an external action completed.
Can it run locally?
Execution, state, and receipt signing can run locally; hosted registry, credential custody, and notary services are optional. The installer, remote skills, registries, and external connectors involve network access.
Is Runx Skill Workflow Runtime free?
Yes. Runx Skill Workflow Runtime is free to use and does not need a paid service.
Is Runx Skill Workflow Runtime open source?
Yes. The source code is on GitHub (runxhq/runx) under the Apache-2.0 licence.
Is Runx Skill Workflow Runtime safe to use?
FollowAgents' FARS review (Nov 13, 2026) scored Runx Skill Workflow Runtime 21/29 on trust and safety. None of the eight safety controls the review checks for was missing outright. The repository has changed since that review.
Is Runx Skill Workflow Runtime still maintained?
Yes. The latest commit on GitHub is from 2026-11-20, within the last 30 days. The latest release is cli-v0.11.1 (2026-11-02).
Who makes Runx Skill Workflow Runtime?
Runx Skill Workflow Runtime is published on GitHub by the account runxhq. FollowAgents has not verified which person or company is behind that account.
What is Runx Skill Workflow Runtime's official website?
The project's GitHub repository lists https://runx.ai as its website. That link is set by the repository owner; the source code itself is on GitHub (runxhq/runx).
Was this page helpful?
View on GitHub ↗ Install ↓

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related concepts