Files
PaperClipAI/packages/paperclip-runner/docs/live-console-protocol-server.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

136 lines
6.2 KiB
Markdown

# Live console Protocol and Demo Server
## Scope
This is the lowest Live console layer. It extends the package-local Codex driver,
typed harness contract, deterministic fixtures, and mock-core demo server. It
does not import or change Paperclip server, UI, database, or production routes.
The browser boundary is HTTP plus server-sent events. The browser sends typed
actions to the demo server. Only the server starts `codex app-server`, reads
the existing Codex login, owns the working directory, and resumes provider
threads. No provider or Paperclip credential is serialized to browser state.
## Deterministic fixture
`protocol/fixtures/codex-driver/driver-conformance.json` records the exact Codex
0.132.0 wire shapes used by conformance tests:
- command, file, permission, user-input, and MCP elicitation requests;
- accept, accept-for-session, decline, cancel, and submitted answers;
- goal get, set, pause, resume, and clear calls;
- subagent parent/child thread source metadata;
- same-turn and stale-turn controls;
- interrupt-before-start and terminal-race outcomes;
- stable reconnect identities and source cursor; and
- hostile diagnostic strings that must be redacted.
The fixture loader rejects missing cases, duplicate request identities,
incomplete goal operations, and incomplete lineage or recovery identity.
## Browser-resolved requests
The driver no longer automatically declines supported runtime requests. It
emits `runtime_request.created`, keeps the JSON-RPC response pending, and waits
for `resolveRuntimeRequest`.
| Upstream request | Canonical kind | Allowed response shape |
| --- | --- | --- |
| `item/commandExecution/requestApproval` | `command_approval` | accept, session accept, decline, cancel |
| `item/fileChange/requestApproval` | `file_approval` | accept, session accept, decline, cancel |
| `item/permissions/requestApproval` | `permission_approval` | empty safe permission grant for turn or session |
| `item/tool/requestUserInput` | `user_input` | typed answer map, decline, cancel |
| `mcpServer/elicitation/request` | `elicitation` | typed content, decline, cancel |
The permission response cannot widen the skillless filesystem or network
boundary. A stale turn, duplicate response, provider cleanup, terminal turn,
protocol failure, close, or reconnect settles the pending request exactly once.
Only the resolution action enters the canonical event; submitted secrets or
answers do not enter the acknowledgement payload.
## Steering and interruption
Successful same-turn steering emits a completed
`steering_acknowledgement` item. A different or child identity is rejected
with `HarnessStaleTurnError` and `stale_turn_rejected`; direct child steering
is never emulated.
An interrupt received while `turn/start` is waiting for a provider turn ID is
queued and sent as soon as the ID arrives. After a terminal fact wins the race,
further steering or interruption returns `already_terminal`. A successful
interrupt request emits an `interrupt_acknowledgement` item without replacing
the session.
## Goals and lineage
The driver probes `thread/goal/get` after create or resume. Success enables
get, set, pause, resume, and clear. A method-not-found response disables goals
and advertises the precise unsupported capability. The installed Codex 0.132.0
binary generated goal bindings but did not enable the method at runtime in the
recorded real session, so the evidence correctly shows `goals: false`.
Root and child threads use `thread.id`, `thread.sessionId`, `forkedFromId`, and
`source.subAgent.thread_spawn` to build lineage. Child start, status, and close
notifications become canonical `thread_lineage` items. Unrelated thread
notifications are ignored instead of being attached to the active session.
## Demo-server API
Start it with:
```sh
pnpm --filter @paperclipai/paperclip-runner demo:live-console -- \
--host 127.0.0.1 --port 4174
```
The transport fails before binding unless `--host` is an explicit loopback
address. The standalone server and Vite middleware use the same admission
guard. It requires loopback local and remote sockets, the exact configured
loopback `Host` plus listening port, same-origin Fetch Metadata, a matching
browser `Origin` on mutations and streams, and `application/json` on every
mutation. Server-side automation uses a random per-launch bearer capability;
it is never returned by an API response or browser state.
Important routes:
- `GET /api/liveConsole/health`
- `POST /api/liveConsole/sessions`
- `GET /api/liveConsole/sessions/:id`
- `GET /api/liveConsole/sessions/:id/events?after=:cursor`
- `GET /api/liveConsole/sessions/:id/stream?after=:cursor`
- `POST /api/liveConsole/sessions/:id/steer`
- `POST /api/liveConsole/sessions/:id/interrupt`
- `POST /api/liveConsole/sessions/:id/requests/:requestId/resolve`
- `POST /api/liveConsole/sessions/:id/goal/:operation`
- `POST /api/liveConsole/sessions/:id/reconnect`
- `POST /api/liveConsole/sessions/:id/close`
The server chooses the working directory. A create body cannot override it.
Provider, Paperclip, cookie, and bearer credential fields in create bodies are
rejected. Runtime-request resolution must match the pending request, its turn,
and the session named by the route; stale, cross-scope, and duplicate responses
fail with `409` before reaching the driver.
Every JSON and event response passes a second bounded redaction layer. Replay
uses the same validated PRP events and reducer as the live view.
The demo keeps at most 16 active sessions and four SSE subscribers per session
by default. Call the close route (the browser does this on reset) to release a
session immediately; server shutdown closes every remaining session and
subscriber. Configuration may lower these limits and is hard-capped at 64
sessions and 16 subscribers.
Residual risk: a loopback service is still reachable by untrusted local
processes, which can reproduce browser headers. The Codex sandbox and approval
policy remain the final execution boundary; this demo transport is not a
multi-user authorization boundary.
## Verification
Run deterministic conformance first, then the real Codex boundary:
```sh
pnpm --filter @paperclipai/paperclip-runner exec vitest run \
src/drivers/codex/codex-app-server-driver.test.ts \
src/mock-core/live-console-demo-server.test.ts
```