v0.9.0 EPIC: Chat-native CodeWhale workrooms for threaded, shareable agent work

[Hmbown]

Problem

CodeWhale should not be only a terminal app or a local web page. For day-to-day agent work, the natural surface is a chat-native workroom: threads, channels, mentions, shareable links, GitHub context, mobile access, and multiple agents or models able to inspect and continue the same work.

The current pieces exist in adjacent form: Runtime API threads/events, the mobile control page, remote bridge docs, Fleet, WhaleFlow, share-link/web UI issues, and remote setup issues. The missing product abstraction is a first-class CodeWhale workroom that all those surfaces share.

This should be CodeWhale-native and provider-agnostic. The goal is not to copy any one chat product. The goal is that CodeWhale work feels like durable collaborative channels over the same Runtime API, with our always-on agent/fleet model underneath.

Product direction

A workroom should make these workflows ordinary:

• A user starts or resumes work from a phone, TUI, web view, or chat bridge.
• Agent work is split into durable channels/threads with stable links.
• Any authorized agent can inspect a workroom link and recover the relevant context, artifacts, approvals, and current state.
• Work can move between GitHub issues/PRs/commits/checks and CodeWhale threads without copy-pasting full transcripts.
• Multiple providers/models can participate; no workroom is locked to one LLM.
• The user can see all active work, direct mentions, thread replies, failures, approvals, and completed receipts in one inbox.

Proposed abstraction

Add a first-class Workroom layer over Runtime API threads/fleet runs:

• WorkroomId, title, workspace/repo/branch identity, owner, members, roles, and visibility.
• WorkroomThread / ExternalThreadRef for channel, DM, issue, PR, CI/check, and mobile/web origins.
• WorkroomEvent over existing thread/fleet/runtime events: message, mention, approval request, worker started, receipt posted, artifact linked, failure, needs-human, resumed.
• WorkroomLink that can be copied into another chat, agent prompt, GitHub issue, or terminal and resolved back to the durable Runtime API state.
• Inbox views: all, DMs, mentions, active threads, needs approval, failed/retryable, completed receipts.
• Artifact refs instead of transcript dumps: fleet logs, receipts, patches, screenshots, test output, release gates.

Acceptance criteria

• A design doc/RFC defines the CodeWhale workroom data model and how it maps to existing Runtime API threads, fleet runs, and external refs.
• Runtime API exposes enough read-only workroom endpoints to list workrooms, list active threads, resolve a workroom link, and fetch bounded state without raw transcript replay.
• At least one existing surface, such as the mobile page or a chat bridge, consumes the workroom projection instead of inventing its own thread list.
• Workroom links can be pasted into a CodeWhale prompt or chat bridge and resolve to scoped context with permissions checked.
• GitHub issue/PR/commit/check refs can be attached as workroom context without copying secrets or full logs into prompts.
• The workroom model supports multiple agents/models and records which agent/model produced each event or receipt.
• Security model is explicit: local-first by default, Runtime API auth required, no public hosted service assumption, no secrets in links, logs, argv, or persisted event payloads.
• Docs explain the relationship between workrooms, Runtime API threads, Fleet, WhaleFlow, chat bridges, mobile, and future web UI.

Non-goals for the first version

• No hosted public CodeWhale cloud service.
• No default-on Slack, Discord, Feishu, Telegram, or GitHub App integration.
• No arbitrary public share links without an explicit operator-controlled auth story.
• No model-specific workroom format.

Related

• v0.8.65 EPIC: Fleet execution substrate for profiled workers #3154 Agent Fleet control plane.
• v0.8.61: Split sub-agents into a headless worker runtime with lightweight TUI projections #3096 headless worker runtime and lightweight projections.
• v0.8.65: Fleet profiles for agent roles, loadouts, permissions, and delegation #3167 Fleet roles/delegation policy.
• codewhale remote-setup wizard: guided cloud + chat-bridge + provider setup (/remote) #2965 remote setup wizard.
• telegram-bridge: serial update dispatch deadlocks approvals — poll loop must never await a turn #2966 and telegram-bridge: streaming & resilience hardening — progress-edit, typing, MarkdownV2, chunking, offset/replay safety, backoff #2967 Telegram bridge robustness.
• #757 Feishu/Lark bot.
• #481 web version and #479 share-link mode.
• v0.8.65: Fleet model classes, loadout auto, and semantic route roles #3205 route-effective model inventory and fleet model selection.

[Hmbown]

Companion issue: #3210 tracks the always-on stateful agent identity that should sit underneath this workroom surface.