Files
PaperClipAI/packages/paperclip-runner/docs/sdk.md
T
Dotta 560e7e48b5 feat(runner): add SDK and developer tooling (#12608)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work.
> - The runner package already provides the production protocol and
execution spine.
> - Contributors still need stable SDK surfaces, deterministic test
tools, and local inspection tools.
> - Those surfaces share generated contracts and must change as one
package boundary.
> - This pull request adds the package-local SDK, labs, examples, and
drift checks.
> - The benefit is a reviewable developer platform that does not change
application execution selection.

## Linked Issues or Issue Description

**Subsystem affected**

`packages/paperclip-runner` — runner SDK, conformance tools, and
developer tooling.

**Problem or motivation**

The production runner spine is present, but package consumers cannot
build deterministic integrations, inspect sessions, or verify
provider-neutral behavior through supported surfaces.

**Proposed solution**

Add browser, React, standalone, live-session, scenario, conformance, and
evaluation surfaces. Add generated contract inventories and
package-local verification scripts. Keep production application routing
unchanged.

**Alternatives considered**

We considered splitting each generated catalog, SDK surface, and demo
into separate pull requests. Those changes share exports, fixtures, and
drift gates. Splitting them would create intermediate package states
that do not build.

**Roadmap alignment**

No overlapping item appears in `ROADMAP.md`. This work extends the
runner package that is already on `master`.

## What Changed

- Add browser, React, standalone, live-session, and issue-thread SDK
surfaces.
- Add deterministic mock control-plane, scenario, conformance, replay,
and evaluation tools.
- Add bounded Codex, OpenCode, and ACPX development transports and
fixtures.
- Keep deferred managed-provider execution fail-closed. Persisted
compatibility data remains readable.
- Add generated capability inventories with their source files and drift
checks.
- Add examples, package documentation, browser checks, and
clean-consumer checks.
- Preserve the reviewed protocol bounds, replay compatibility aliases,
process environment isolation, and semantic redaction limits.
- Update the ACPX package patch that the existing workspace patch
registry already tracks.
- Do not change `pnpm-lock.yaml`, repository workflows, server runtime
selection, or the application UI.

## Verification

GitHub Actions is the verification authority for this pull request. The
repository CI, package TypeScript and Rust checks, package tests,
generated-output drift checks, browser checks, security scans, and
Greptile review must pass on the exact head.

Local test suites were not run because this series uses parallel GitHub
Actions for verification.

## Risks

This is a large greenfield package change. The main risks are public
export drift, generated-output drift, and optional React consumer
compatibility. Package boundary checks, clean-consumer checks, and
browser tests cover those risks. Production adapter selection and server
execution are outside this pull request.

## Stack

1. **This PR:** runner SDK and developer tooling.
2. [Codex production server
integration](https://github.com/paperclipai/paperclip/pull/12616).
3. [Provider-neutral task-thread
UI](https://github.com/paperclipai/paperclip/pull/12617).

## Model Used

OpenAI Codex, GPT-5, high-reasoning mode, with tool use and code
execution.

## Checklist

- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have described the issue in-PR following the feature request
template
- [x] I have not referenced internal/instance-local Paperclip issues or
links
- [x] My branch name describes the change and contains no internal
Paperclip ticket id
- [ ] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented risks above
- [ ] All Paperclip CI gates are green
- [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge
2026-08-31 21:33:11 -05:00

7.6 KiB

SDK Browser SDK and Reference Console

Public package surface

SDK freezes the browser SDK as package version 0.1.2.

Version 0.1.1 adds direct chat mode. The browser sends the operator message as plain Codex input. The protocol inspector still shows the event stream, reducer state, requests, and replay data. The Codex task fixtures remain available from the session controls.

Version 0.1.2 adds the canonical events for each transcript item to the optional debugEvents projection. ToolItem renders these events inside a nested Debug details disclosure. Completed Terminal rows stay folded by default, while operators can inspect command input, output, provider items, delta updates, event ids, sequence numbers, and timestamps. Long chat and command content wraps at mobile widths without horizontal page scrolling.

Import Purpose
@paperclipai/paperclip-runner Protocol, reducer, and existing package contracts
@paperclipai/paperclip-runner/browser Framework-free HTTP/SSE client, protocol types, transcript projection
@paperclipai/paperclip-runner/react Hook, reference console, and reusable React components
@paperclipai/paperclip-runner/styles.css Scoped light-theme token and component styles

React and React DOM are peer dependencies. The extracted surface adds no runtime dependency. It is client-only: it uses window, sessionStorage, Fetch, and EventSource, so applications must mount it in the browser rather than render it on the server.

Framework-free client

import { createRunnerClient } from "@paperclipai/paperclip-runner/browser";

const client = createRunnerClient({ baseUrl: "/api/liveConsole" });
const manifests = await client.fetchManifests();

RunnerClient exposes manifest listing; session create/read/close; event history and SSE streaming; turn start, steering, and interrupt; typed request resolution; goal operations; and reconnect. RunnerClientError preserves the HTTP status, stable error code, and redacted server message.

The protocol server remains authoritative. The browser client never creates a second event model, invents capabilities, changes run/session/provider identities, or removes duplicate events before the shared reducer sees them.

React hook

import { useRunnerConsole } from "@paperclipai/paperclip-runner/react";

function Console() {
  const runner = useRunnerConsole({ baseUrl: "/api/liveConsole" });
  return <p>{runner.connection}: {runner.replayParity ? "match" : "waiting"}</p>;
}

useRunnerConsole returns manifests and selection, canonical server state, the exact event list, reducer snapshot, transcript projection, connection and retry state, composer state, steering acknowledgements, a single polite announcement channel, errors, busy state, lineage selection, replay controls, and verbs for every supported mutation. Durable history is re-read before a reconnected stream opens. Replay uses the same reducer as live state.

Components

The ./react export includes:

  • primitives: Button, Badge, Card, Textarea, Tabs, Menu, Dialog, Tooltip, and Banner;
  • protocol views: Conversation, Message, ReasoningItem, ToolItem, RequestCard, SessionTimeline, Inspector, ReplayControls, and ConnectionBanner;
  • composition: Composer, RunnerConsoleApp, and useRunnerConsole.

Components take reducer/protocol shapes. Every component owns a stable data-slot and pcr- class. Terminal, file, tool, plan, request, failure, lineage, goal, connection, and replay states come only from canonical events or public session state.

The reference server has two driver modes behind the same HTTP/SSE contract. Deterministic manifests use LiveConsoleScriptedDriver. Real chat mode uses CodexAppServerDriver, which starts a real local codex app-server process. The Node server owns provider authentication and the disposable working directory; browser code receives only canonical runner events.

Five extension points

These are the complete extension surface for 0.1.2:

  1. Conversation, Message, ReasoningItem, and ToolItem accept renderItemBody(item) for markdown, highlighting, or another item body.
  2. RequestCard accepts renderRequestDetail(request) for its detail region.
  3. Composer accepts leadingActions and trailingActions.
  4. Consumers may override --pcr-* properties under .pcr-root.
  5. createRunnerClient and useRunnerConsole accept baseUrl, fetchImpl, and eventSourceFactory transport injection.

There is no headless distribution, component registry, render-prop shell, dark theme, toast layer, or virtualized transcript in this version.

Transport and credential boundary

const runner = useRunnerConsole({
  baseUrl: "/runner",
  fetchImpl: (input, init) => fetch(input, {
    ...init,
    headers: { ...Object.fromEntries(new Headers(init?.headers)), "x-app-session": appSession },
  }),
  eventSourceFactory: (url) => new EventSource(url),
});

Transport injection is for browser-to-protocol-server authentication, proxies, and test doubles. Provider credentials stay on that server. Do not place a provider bearer token in component props, manifest text, events, diagnostics, browser storage, or an EventSource query string. The package demo fixes and validates its workspace server-side and redacts provider paths and credentials.

Tokens and contrast

styles.css declares a light-only color-scheme and all visual values under .pcr-root. The shipped foreground/surface pairs are:

Foreground Surface Contrast
--pcr-foreground --pcr-background 16.4:1
--pcr-card-foreground --pcr-card 17.6:1
--pcr-primary-foreground --pcr-primary 15.6:1
--pcr-muted-foreground --pcr-muted 5.0:1
--pcr-success --pcr-success-surface 6.5:1
--pcr-warning --pcr-warning-surface 5.4:1
--pcr-danger --pcr-danger-surface 6.2:1
--pcr-accent --pcr-accent-surface 5.5:1

When a consumer overrides either side of a pair, that consumer owns a new contrast check. Keep normal text at 4.5:1 or better. Motion respects prefers-reduced-motion. All mobile controls use the --pcr-touch-target minimum.

Minimum supported widths are 320px for Conversation, Composer, and Inspector, and 280px for RequestCard. RunnerConsoleApp renders one responsive tree: three panes above 900px and one selected pane below it.

Keyboard and accessibility contract

  • The transcript is a labelled role="log"; consumers render the hook's announcement once in an aria-live="polite" region.
  • Composer Enter sends or steers; Shift+Enter adds a line. Stop remains a separate button.
  • Tabs use arrow-key roving focus plus Home/End.
  • Menu supports Enter/Space/ArrowDown, arrows, Home/End, and Escape with focus restoration.
  • The native dialog traps focus, closes on Escape, and restores its trigger.
  • Reasoning and tool disclosures use native summary keyboard behavior.
  • Replay Left/Right steps; Space toggles play from the scrubber.
  • Status is always named in text, not communicated by color alone.
  • Informational banners do not steal focus. A blocking reconnect failure may.

Reference applications

examples/reference-console/ composes RunnerConsoleApp only from public exports. examples/mini-consumer/ imports only the three public SDK subpaths and visibly exercises every extension point. Both run against the package fake driver or the real Codex driver through the same server-only adapter.

See the hand-run tutorial and the implementation decision record.