For AI agents: the complete documentation index is available at https://a3s-lab.github.io/ash/en/llms.txt, the full documentation bundle is available at https://a3s-lab.github.io/ash/en/llms-full.txt, and this page is available as Markdown at https://a3s-lab.github.io/ash/en/guide/index.md.
ash
  • English
  • next

      • Get started with ash

      ash is an AI Native Shell for Coding Agents. It accepts typed ASH/1 requests, runs them across bounded Tokio I/O and Rayon CPU planes, and returns canonical ASON with retrievable evidence instead of terminal text formatted for people.

      Current status

      The next docs track main. Source, cross-platform installers, and the six-target release workflow are implemented, but release credentials are not provisioned and no supported signed binary is published.

      Choose your path

      You want to…Start hereThen read
      Use ash in a Coding AgentCoding Agent integrationComplete capability map
      Build a persistent Agent HarnessASH/1 protocolASON format and security
      Understand the runtimeArchitectureBenchmarks and evidence
      Install or operate releasesInstallationCLI and security

      Choose an entry point

      ScenarioEntry pointContract
      Long-lived Agent Harnessash rpcFramed handshake, concurrent requests, cancellation, permits, and result references
      One typed callash runOne bare ASON request from standard input
      ASON authoringash asonValidation plus the unique canonical representation
      Binary identityash --build-infoVersion, target, feature profile, and protocol identity
      Install and lifecycleinstall.sh / install.ps1 / ash self …Verification, transactional activation, status, update, recovery, rollback, and uninstall

      Run a canonical request

      Create request.ason to search src for literal TODO:

      t:1
i:17
o:g
a{q,p,f}:
TODO,[src],0
u{tok,rec,ms}:
256,64,30000
      ash run < request.ason

      The envelope order is exact: message type, positive request ID, operation, typed arguments, then token/record/time budgets. The response is canonical ASON as well. Errors contain stable diagnostic codes and typed fields—not color, progress bars, or prose suggestions.

      How one request flows

      1. Decode canonical ASON into typed ASH/1 semantic IR.
      2. Validate capabilities, budgets, paths, digests, and graph structure.
      3. Schedule I/O on Tokio and splittable CPU work on Rayon under one governor.
      4. Stable-merge concurrent results and retain complete evidence when needed.
      5. Emit the smallest honest ASON projection; follow its reference for more.

      The complete capability map lists all fifteen operations, runtime guarantees, transaction recovery, output storage, delivery, test evidence, and deliberate non-goals. The checked-in Coding Agent Skill turns those contracts into a reusable repository workflow.