Skip to content

docs

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

docs

README.md

Agentic Harness Docs

Long-form documentation for Agentic Harness. Use the top-level README for install, examples, and the common command path. Use this folder for architecture, runtime protocols, deployment boundaries, feature status, and release checks.

Recommended Reading Order

  1. Execution Targets: choose local, CI, remote sandbox, Node host, or Cloudflare boundary output.
  2. Runtime Config: configure models and providers.
  3. HTTP SessionEnv Protocol: connect remote sandboxes.
  4. Feature Status: verify what is implemented today.

Architecture

  • Execution Targets — local development, CI, remote Linux sandboxes (Vercel Sandbox, Daytona, E2B), and Cloudflare edge: what runs where, and why.
  • Runtime Config — provider defaults, env-key resolution, OpenAI-compatible model registration, role/call/default precedence, and runtime.json discovery rules.
  • HTTP SessionEnv Protocol — the JSON-over-HTTP wire format that lets a Rust agent run shell, file, and env operations inside any remote sandbox that speaks the protocol.
  • Sandbox Connectors — when to use the built-in SandboxConnector helpers and when to generate a project-local connector.
  • Virtual Sandbox — hostless in-memory filesystem and small shell subset for tests, support agents, and Worker-style adapters.
  • Cloudflare Runtime — Worker boundary build pipeline, Durable Object bindings, the WASM JSON ABI, and the --worker-app / --worker-wasm / --worker-wasm-crate build options.

Deployment Guides

  • Node Hosts — package a native Rust agent behind node server.mjs for hosts that require a Node entrypoint.
  • Cloudflare Workers — generate Worker boundary files and link a Worker-compatible app adapter.
  • GitHub Actions — run agents in checked-out repositories for issue triage, PR review, and release jobs.
  • GitLab CI — run agents in GitLab runners with explicit CI credentials.

Status & Releases

  • Feature Status — capability matrix with code, tests, and docs evidence for every major surface.
  • HarnessScore — documented in the top-level README and feature status; use agentic-harness score --workspace . --run latest to evaluate run artifacts and expose latestHarnessScore in dashboard JSON.
  • Roadmap — the next product slice and explicit non-goals.
  • Release Smoke Test — clean-machine pre-publish checklist (smoke --json, release-check --json, install script, Homebrew formula, changelog).