0.30.0

​

Upgrading

bunx smthrs update        # upgrade the CLI
bunx smthrs upgrade       # agent-assisted: reads the changelogs and applies what your project needs
bunx smthrs packs update  # refresh installed workflow packs

​

A UI kit for the interfaces your agents build you

import { Message, Reasoning, ReasoningSummary, ToolCall, MessageResponse, Sources } from "smthrs/ui";

<Message role="assistant">
  <Reasoning duration={12}>
    <ReasoningSummary text={reasoningSummary} />
  </Reasoning>
  <ToolCall name="edit_file" state="approval-requested" args={{ path }} approvalSlot={<ApproveDeny />} />
  <MessageResponse content={markdown} />
  <Sources sources={sources} />
</Message>

​

Live widgets for custom run UIs

• NodeChatStream: one agent’s live chat, streaming as it works.
• FleetTable and NodeStageStrip: a dashboard for fan-out runs (many nodes, same task shape) instead of a flat event list.
• RunEventLog: now structured, severity-tinted rows, coalesced heartbeats, and no raw JSON.
• MonitorButton: one-click deep link from your UI into the Monitor.

import { NodeChatStream, FleetTable, MonitorButton } from "smthrs/gateway-ui";

<NodeStageStrip runId={runId} stages={stages} />
<NodeChatStream runId={runId} nodeId="implement" subtitle="opencode · kimi-k3" />
<FleetTable runId={runId} items={repos} />
<MonitorButton runId={runId} />

​

A triage-first Monitor

• Needs you: pending approvals, human-parked runs, and failures from the last 24h, with approve/deny inline.
• Active now: what is running this second.
• Operational stats move to a quiet footer; frames and the raw XML tree tuck into a Debug fold.
• A keyboard kit for mouse-free triage: / to search, j/k to move, Enter to open, Esc to back out.

​

Charts and a workflow canvas

import { ChartContainer, chartConfig } from "smthrs/ui/adapters/chart";

<ChartContainer id="throughput" config={chartConfig([{ key: "landed", label: "Tasks landed" }])}>
  <AreaChart data={data}>…</AreaChart>
</ChartContainer>

​

smithers oneshot

bunx smthrs oneshot "add a loading spinner to the login button"

​

A hardened browser viewer

​

Model registry: Claude orchestrates, Kimi K3 joins

​

Community thanks

​

Other improvements

• Detached runs fail fast instead of fake-succeeding. up -d now validates the workflow before spawning: a bad file prints file:line:col and exits non-zero with no run created, instead of printing a success run id for a run that never started.
• Runs carry launch attribution. Who or what started a run (a CLI call, a plugin session, a gateway client) is persisted and surfaced by getRun/listRuns.
• A friendlier gateway-client offline mode. An unreachable gateway is classified as GATEWAY_UNAVAILABLE, keeps last-known rows, and logs once per session instead of spamming console.error.
• <Memory> via createSmithers resolves a real memory runtime and store end-to-end, with real-backend e2e coverage.
• listRuns accepts an offset for server-side pagination.
• 106 fix commits landed overall, three of them security fixes; 257 documentation pages were rewritten for concision with mechanical checks that nothing was lost.

​

Upgrade notes

• Default orchestration and gating now route to Claude if you use the shipped registry defaults rather than a custom agents.ts.
• The kimi oneshot seat now runs Kimi K3, not K2.7.
• A detached run can now fail fast before it starts. Tooling that parses up -d output expecting an immediate run id will now see a non-zero exit and a file:line:col error when the workflow file is invalid.
• gateway-client no longer logs every failed RPC when the gateway is unreachable; it logs once per session and surfaces GATEWAY_UNAVAILABLE.
• listRuns rejects a malformed offset with INVALID_REQUEST instead of ignoring it. Callers that omit it are unaffected.

​

The full changelog