Skip to content

coderunner.md

Latest commit

 

History

History
126 lines (81 loc) · 4.07 KB

coderunner.md

File metadata and controls

126 lines (81 loc) · 4.07 KB
Raw

CodeRunner

"A local sandbox for your AI agents."
Website https://github.com/instavm/coderunner
GitHub https://github.com/instavm/coderunner
Stars GitHub Stars
Classification agent-native
Category Category README

Official Website

https://github.com/instavm/coderunner

Official Repo

https://github.com/instavm/coderunner

How to Use (Agent Onboarding)

Install CodeRunner from GitHub and run agent workloads in isolated Apple containers

The entry was added as a recent high-signal candidate because the official materials position it around AI agents and expose machine-to-machine interfaces rather than a human-only UI.

Agent Skills

Status: ⚠️ No dedicated catalog skill verified in this entry. Search ClawHub or the upstream README for the latest installable skills.

MCP

Status: ✅ Verified/documented by upstream or MCP server listings.

What It Does

CodeRunner provides a local sandbox for your AI agents. Its agent-facing value is to give autonomous systems a programmatic primitive that can be invoked during a task loop without requiring a human to operate a dashboard for each action.

Why It Is Agent-Native

Criterion Evidence
Agent-first positioning Upstream tagline/README explicitly targets AI agents: "A local sandbox for your AI agents."
Agent-specific primitives Apple Containers, local sandbox CLI, MCP.
Autonomy-compatible control plane Agents can call the CLI/API/MCP surface repeatedly during a run after operator provisioning.
M2M integration surface Apple Containers, local sandbox CLI, MCP.
Identity / delegation Delegation is enforced through the runtime boundary, policy configuration, or caller credentials; auditability is attributable to the configured agent session.

Primary Primitives

Primitive Description
Agent runtime surface Apple Containers, local sandbox CLI, MCP
Session boundary Isolates work by run, process, sandbox, memory namespace, or policy scope depending on deployment
Machine-readable output Returns artifacts, traces, memory records, or policy decisions for direct agent consumption

Autonomy Model

Operator provisions CodeRunner
  ↓
Agent connects through the documented CLI/API/MCP surface
  ↓
Agent performs bounded actions under policy/session limits
  ↓
Results, state, traces, or approvals return to the agent loop

Identity and Delegation Model

  • Agent session identity is represented by the caller configuration, sandbox/session, memory namespace, or policy subject.
  • Delegation controls come from runtime limits, allowlists, approval thresholds, or deployment credentials.
  • Audit trail is provided by logs, traces, memory history, policy events, or sandbox artifacts exposed by the service.

Protocol Surface

Interface Detail
Primary surface Apple Containers, local sandbox CLI, MCP
Repository https://github.com/instavm/coderunner
MCP ✅

Human-in-the-Loop Support

Humans configure the service and policy boundaries. Normal agent operation proceeds unattended until a deployment-specific limit, approval gate, or error condition requires review.

Why Generic Alternatives Do Not Qualify

Alternative Why It Fails
Generic developer tool Does not expose an agent-first primitive or machine-to-machine loop tailored for autonomous agents.
Manual dashboard workflow Requires a human to click through the same action the agent needs to perform programmatically.

Use Cases

  • Autonomous coding, research, support, or operations agents that need a local sandbox for your AI agents
  • Multi-agent systems that need shared, auditable infrastructure boundaries
  • Production agent deployments where repeatable machine interfaces matter more than human UI