Files
PaperClipAI/packages/paperclip-runner/docs/tutorials/live-console.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

7.3 KiB

Live console: Run the Live Codex Protocol Console

What this is

This tutorial runs the Live console browser console. The protocol demo server tutorial used curl. This one uses the real user interface.

What this proves

The console shows one live session. You can chat with it, steer it, stop it, answer its requests, change its goal, watch its child threads, break its connection, refresh the page, and replay the whole record. Every screen reads the canonical protocol. Nothing in the browser holds a provider login.

Before you start

Run every command from the repository root. You do not need a Codex account for this tutorial. The default driver replays deterministic demo chats.

Step 1: Run the deterministic checks

pnpm --filter @paperclipai/paperclip-runner exec vitest run \
  src/mock-core/live-console-scripted-driver.test.ts \
  src/mock-core/live-console-demo-server.test.ts \
  devtools/browser/src/live/transcript-model.test.ts

Expected result: all three files pass. They cover the demo driver, the server routes, and the pure transcript model the console renders.

Step 2: Start the console

pnpm --filter @paperclipai/paperclip-runner console:live-console

Keep this terminal open. Open http://127.0.0.1:4180/ in a browser and press Live console.

Expected screen: an empty transcript that says Pick a demo chat or start a blank session, a list of demo chats on the left, and a disabled Send button.

Step 3: Run one clean turn

  1. Press Completion in the demo chat list.
  2. Press Run this demo chat.

Expected facts:

  • The connection badge says connected.
  • Your own message appears first.
  • A collapsed Reasoning block appears and opens while it streams.
  • A Tool block shows its input and then its output.
  • The answer streams into view with a blinking cursor.
  • The turn ends with a Turn completed badge.

Step 4: Steer a running turn

  1. Press Same-turn steering, then Run this demo chat.
  2. While the turn runs, the primary button says Steer.
  3. Type Keep it to one sentence. and press Enter.

Expected facts:

  • A steering chip appears as pending and then becomes acknowledged.
  • The chip never merges into the answer before the acknowledgement.

Now type text while the turn is running, wait for Turn completed, and press the button. The chip becomes rejected — turn already ended and offers Send as new message. Your text is never discarded.

Step 5: Stop a turn in three ways

Run each demo chat and press Stop:

Demo chat When to press Stop Expected result
Interrupt before start right away Cancelled before start, and no answer item is created
Interrupt during generation while text streams the stream stops, the partial text stays, and an interrupted divider appears
Interrupt during a tool call while the tool runs the tool keeps its last status and shows no result

The session is never replaced. After a stop you can simply send again.

Step 6: Answer requests

  1. Press Command and file approvals, then Run this demo chat.

Expected facts:

  • A blue banner above the composer says 1 request waiting — Review.
  • The command card offers Approve, Approve for session, Reject, and Cancel, because the request offers those four actions.
  • After you press Approve, the whole action row locks until the resolved event arrives. A second click cannot answer twice.
  • The resolved card collapses to one line that names the action you chose.
  • The next card offers only Approve and Reject, because that request offers only two actions.
  1. Press User input and expiry, then Run this demo chat.

Type an answer and press Submit. Leave the second request alone. After a few seconds it becomes expired before response and stays in the record.

Step 7: Change the goal

  1. Press Goal lifecycle, then Run this demo chat.
  2. Press Goal, then Set goal…, type an objective, and confirm.
  3. Press Goal, then Pause.

Expected facts: the banner shows the objective, the state, and the time of the last change. Resume is disabled while the goal is active, and Pause is disabled while it is paused.

Now press Goals unsupported and run it. The Goal button is disabled, not hidden, and it names the exact upstream reason: Goal operations unsupported: app-server 0.132.0 does not advertise the goals capability. Open the Protocol inspector and the Capabilities tab. The same string appears there.

Step 8: Watch child threads

Press Parent and child threads, then Run this demo chat.

Expected facts:

  • The Threads panel lists the root session and two child threads.
  • One child ends Completed and one ends Failed. Both stay in the tree.
  • Selecting a child shows a breadcrumb and a disabled composer that says Direct steering of child threads is not supported by this app-server.

Step 9: Break the connection and recover

  1. Run any demo chat and wait for it to finish.
  2. Press Simulate connection loss.

Expected facts: a banner says the connection was lost, and the composer is disabled.

  1. Press Retry now.

Expected facts: the badge returns to connected and the transcript is exactly what it was before the drop. Nothing is lost and nothing is duplicated.

Now reload the page and press Live console again. The same session returns with its full transcript.

Step 10: Replay the record

Press Replay.

Expected facts: a REPLAY badge appears, the transcript resets, and a stepper appears. Press Step forward to walk the session forward one event at a time. Press Play to run it. Press Exit replay to return to live.

Step 11: Inspect the protocol

Press Protocol inspector.

  • Events: every canonical event with its sequence, type, and full payload. Filter by text or by event type.
  • Requests: every runtime request with its lifecycle.
  • Capabilities: what this app-server supports, with a reason for each unsupported row.
  • Session: identities, cursors, gap state, and the live-versus-replay reducer comparison. It should read match.

Check the Credentials in browser row. It must read no.

Step 12: Check the keyboard

  • Tab reaches every control, including the inspector and the dialogs.
  • Enter sends. Shift and Enter add a line.
  • Escape while a turn runs moves focus to Stop. It never stops the turn on its own.
  • Arrow keys move between inspector tabs and between threads.
  • End jumps the transcript to the newest item.

Step 13: Use the real Codex driver

This step needs a working Codex login on the machine.

PAPERCLIP_LIVE_CONSOLE_DRIVER=codex \
  pnpm --filter @paperclipai/paperclip-runner console:live-console

The routes and the screens are the same. The demo chat list still appears, but the manifest only supplies the first message; the real model produces the rest. The Codex login stays on the server.

Step 14: Clean up

Stop the console with Ctrl+C. The server-owned working directory is a temporary directory and is not part of your repository.

Where to go next