Dudu-0223
ff563d07e0
Models coordinate subagents through send_message, interrupt_agent, list_agents,
and Team task owners, but every one of those addressed a member by a
36-character Session id. Each new child now has a name, unique among its
creator's children, recorded as a field of the subagent/catalog entry the
creator already appends (payload version 3).
The Session id stays the only execution identity for routing, authorization,
persistence, task ownership, and navigation. A name is a model-facing
reference: 1-32 lower-case letters, digits, or hyphens starting with a letter,
never `lead`, so a UUID never parses as one. Omission selects the first free
`agent-N`; an explicit duplicate or invalid name fails before startup. The
creator reserves the name synchronously at admission and releases it when the
catalog entry commits or startup fails, so no second write, rollback protocol,
or optional naming service exists.
The root is `lead`, and a descendant's full name is its path from the root
without a `lead/` prefix, such as `reviewer/checker`. Sibling-unique segments
make full names tree-unique without a root registry, and cold resume rebuilds
a member's path from the catalogs along its creation chain. A member below an
unnamed historical creator keeps its Session id as its name. resolveName()
reads only the catalogs along the path and grants no authority.
Tools, runtime-authored text, and the Web client consume names: delegation
takes an optional name and returns { kind, subagentId, name }; send_message and
interrupt_agent take `agent`; list_agents rows carry name, status, and
description; Team task results report the owner's name while storage keeps
ownerId; identity runtime context, message framing, and settlement notices use
full names. Recorded sessions, CLI and Web expectations, the Python runtime
snapshots, bilingual docs, an upgrade guide, and an Agent Note follow.
description, kind
| description | kind |
|---|---|
| The experimental group map: publicly installable pre-stable prototypes. | package-group |
packages/experimental
English | 中文
Summary
Install and compose experimental capabilities explicitly, without a product-support commitment. Package naming and publication follow the experimental package policy. The optional-bundle rule determines which capabilities ship switched off with the installation for explicit selection in the GUI plugin manager’s Official group; the Liveshare owner bundle and its shared protocol package are among them. The plugin manager page owns which bundles appear by default and how to enable hidden ones.
Table of Contents
Experimental, optional, and Official
Experimental means the complete public capability remains under evaluation and is not a product-support commitment. Its behavior or API may change, and the capability may be withdrawn; engineering, security, documentation, testing, and released-data obligations still apply. A package belongs here only when its entire public contract is experimental or internal-only; an experimental option inside a product package stays with that product role.
Optional means a user explicitly selects the capability. Official means the project maintains and lists the package in the Plugins page. Neither describes maturity; publication, default installation, and GUI visibility do not determine maturity either. Product packages outside this group have maintained product roles, while their public APIs still follow the repository's pre-stable policy.
Promotion requires a named product role, documented public behavior and limitations, current consumers, release contents, and test evidence. Move the package to its product group, apply that group's naming rules, and update affected consumers together. The subtree rules govern dependency isolation and publication.
Packages
| Package | Role | ctx key |
|---|---|---|
action-cards |
Optional action cards below answers: the proposal tool, card state, user actions and status notices | ctx.actionCards |
client-ui-action-cards |
Card frame below completed answers, kind bodies and the proposal process row | — |
creation-recommendations |
Explicit-need creation of persistent removable interface plugins | — |
client-ui-creation-recommendations |
Personalized page card body with location selection and undo | — |
plugin-recommendations |
Explicit-need official plugin recommendations in conversations | ctx.pluginRecommendations |
client-ui-plugin-recommendations |
Official plugin card body with catalog text and enabling | — |
personalized-recommendations |
Default-disabled bundle of personalized pages, plugin recommendations and their action cards | — |
tool-session-query |
Workspace-authorized model session search, trace, and event reads | — |
session-search |
Optional experimental global session search tools | — |
ralph-bundle |
Optional repeated delegation with an isolated workflow engine | — |
terminal-bundle |
Optional global persistent terminal tools | — |
badge-skill-bundle |
Optional powered-by-dsh badge skill | — |
session-titles-bundle |
Optional titles following human prompts in a conversation | — |
hook-protocol |
Hook bridge wire types and durable events | — |
hooks-claude-code |
Claude Code hook bridge | — |
hooks-codex |
Codex hook bridge | — |
webhook |
Authenticated delivery dispatch and Workspace Session creation | ctx.webhookRuntime |
webhook-github |
Signed GitHub webhook ingress | — |
session-title-all-prompts-llm |
Session titles from all human prompts | — |
tool-terminal |
Persistent terminal tools | — |
tool-ralph |
Bounded repeated delegated tasks | — |
skill-badge |
The powered-by-dsh badge skill | — |
translator |
Anonymous Google and Bing text translation | ctx.translator |
client-ui-cot-translation |
Expanded reasoning translation and original-text controls | ctx.cotTranslation |
cot-translation-bundle |
Default-disabled reasoning translation composition | — |
speech-to-text |
Named speech recognition providers | ctx.speechToText |
speech-to-text-sensevoice |
Managed local SenseVoice inference | — |
api-speech-to-text |
Authenticated transient transcription Remote | ctx.speechController |
client-ui-voice-input |
Microphone capture and guarded draft insertion | — |
plugin-market-bundle |
Default-disabled optional custom plugin indexes: the index service and the Discover tab | — |
liveshare |
Opt-in LiveShare host bundle: Relay-backed Session sharing at the view or write tier, with forking and attachment uploads as independent switches |
injects connection; serves the viewer page through the Host's web resources |
liveshare-protocol |
Wire frames and the identity contract shared by the owner bundle and the Relay | library and protocol — no ctx key |
voice-input-bundle |
Default-disabled optional voice input composition | — |
client-ui-stats |
Per-turn and per-step context, timing, cache, and tool cost for Web | — |
auto-review |
Explicit Web layer for same-model review before each native or PTC inner tool call | — |
subagent-oar |
Run standalone external coding-agent tasks through OAR | ctx.subagents |
tool-subagent-oar |
Select a configured external agent for each delegated task | ctx.tools |
claude-code-mods |
Run Claude Code mods as plugins: their hook chains on harness extension points and a band above the prompt | ctx.claudeCodeMods |
client-ui-claude-code-mods |
The Web band that draws a mod's tree above the prompt and sends button clicks back | — |
ptc-runtime-python |
CPython subprocess backend for the PTC execution seam | ctx.ptcRuntime |
| CUA | Application/window objects, authorized execution, and permission controls. | ctx.cuaAccess |
| macOS CUA | Swift native execution and platform preparation. | ctx.computerUse |
cua-bundle |
Optional Computer Use switch for PTC sessions | — |
computer-use-cua-driver-mcp |
Use an installed Cua Driver through MCP | ctx.computerUse |
computer-use-cua-driver-native |
Embed the Cua Driver native npm runtime | ctx.computerUse |
browser-use-playwright-mcp |
Playwright browser tools over MCP | ctx.browserUse |
browser-use-chrome-devtools-mcp |
Chrome DevTools inspection and browser control over MCP | ctx.browserUse |
browser-use-stagehand-native |
Stagehand browser operations with explicitly configured native models | ctx.browserUse |
browser-use-runtime |
Session-owned browser resources shared by experimental providers | — |
inspector |
Cross-realm CDP hub for Host debugging, Client Runtime inspection, network capture, and Cordis trees | ctx.inspector |
session-inspector |
Sidebar tables for raw Session logs and Chat nodes | — |
inspector-profile |
Optional Web bundle for Session log and Chat node inspection | — |
tool-worktree |
Optional bundle and model tool for creating and entering worktrees | ctx.tools |
webworker-packer |
Builds the gzip-compressed VFS image consumed by the browser worker preview | library and CLI — no ctx key |
webworker-runtime |
Runs the harness plugin tree inside a dedicated browser worker | library and worker entry — no ctx key |
worktree |
Creates a named Git worktree and changes the calling Session directory | ctx.worktrees |
forge |
Code host provider registry with shared change request polling | ctx.forge |
forge-github |
GitHub provider that reads pull requests through the host GitHub CLI | registers a provider on ctx.forge |
forge-session |
Binds change requests to the Sessions that created, pushed, or tracked them | ctx.forgeSession |
client-ui-forge |
GUI panel for a Session's Git checkout, worktrees, and bound pull requests | ctx.forgePanel |
forge-bundle |
Default-disabled pull request and worktree panel composition | — |
Related documentation
- Experimental publication reference — public defaults and private exceptions.
- Worktrees — explicit branch and checkout creation.
- Forges — change request providers, status polling, Session bindings, and the panel.
- Computer use — desktop provider choices.
- Browser use — browser provider choices and Session ownership.
- Agent Teams subsystem — durable Team types and the
ctx.agentTeamsservice API. - Experimental subtree rules — what experimental status does and does not relax.
Dev Note
Working context for maintainers — click to expand
None.