[wayfinder:map] OpenBot v1

[lukemaj]

State: active wayfinder map. Not directly dispatchable.

Current navigation, 2026-08-30

This map is historical context plus current routing. Do not dispatch from the original frontier below.

Fresh Gate 1 accounting:

• 64 open issues
• 35 ready-for-human
• 29 explicitly blocked, frozen/unpublished, human-decision, post-v1, or umbrella/map issues
• 0 agent-running
• 0 ready-for-agent
• every open issue starts with an explicit State: line
• BH-01 through BH-37 are mapped without a missing or extra identifier

Immediate ordered delivery path:

1. Human action-time authorization for corrected [task] Converge familiar Messenger and isolated-runtime trains #116 exact-head connected-PWA live acceptance.
2. Human disposition of PR feat: converge familiar and isolated v1 trains #136.
3. Rebuild [bug] Keep selected-Bot Instructions inside its panel #137, then [bug] Give the phone Chat composer full touch targets #139, on the accepted corrected train. Do not merge stale PRs fix: contain selected-Bot Instructions layout #138 or fix: give the phone Chat composer full touch targets #140.
4. Run fresh exact-candidate [task] V1 messenger acceptance through Computer Use #80 acceptance.
5. Rebuild, live-prove, and publish [task] Honest Isolated config and Host config contract #83.
6. Recheck [task] Disconnect Host grants from v1 authorization #84 and complete its focused real-provider proof.
7. Resolve the [task] Give new Sessions the mode-aware OpenBot environment brief #85 human decision, then implement [task] Give new Sessions the mode-aware OpenBot environment brief #85 and serialize [task] Version or migrate runtime OPENBOT.md seed #70/[task] Keep the OpenBot environment brief correct across Session lifecycle #86.
8. Run [task] V1 Screen-first config acceptance through Computer Use #87, then final integrated [task] Run adversarial runtime regression acceptance #99 acceptance.

Separate human gates:

• [task] Spawn Grok Build over ACP #34: rebuilt final candidate plus explicit authority for one real Grok provider Turn that may consume account usage
• [task] Screenshot Card and attach an image #41: approve or reject the minimal v1 image-attachment contract
• [task] Give new Sessions the mode-aware OpenBot environment brief #85: approve or reject the portable Session-brief seam
• [grilling] Talk voice: short replies and screenshots #21: closed voice contract, with inspection deferred until the human explicitly requests the pause

Current PR topology is exactly #136, #138, and #140. PR #136 is the corrected combined candidate at 94af08e7a4dc571593dba5dc694ca12aebb3c65e. PRs #138 and #140 are frozen predecessor evidence and have no automatic issue-closing references.

Detailed current evidence:

• Gate 1 census: [wayfinder:map] OpenBot v1 #1 (comment)
• original objective contract audit: [wayfinder:map] OpenBot v1 #1 (comment)
• [task] Converge familiar Messenger and isolated-runtime trains #116 live authorization packet: [task] Converge familiar Messenger and isolated-runtime trains #116 (comment)

Historical map

The original map below is preserved as decision history. Its Frontier and Not yet specified sections are not current executable order.

Chart of decision tickets for OpenBot v1.

Destination

A cloneable MIT OpenBot: one host command starts Screen (compose) and is the ACP parent, Cloudflare Tunnel, one Password, PWA messenger with Eyes, N Bots with private Screens on one Computer, PinchTab, user-picked host Harness (PATH only), Takeover for 2FA.

Talk should feel like Grok Bot: short human replies, a screenshot when it helps. The Harness is a real coding CLI already on the host OS (Codex, Claude Code, Grok Build, Kimi), not a chat stack we invent, and not a process inside XFCE.

Locked (grilling)

See CONTEXT.md and docs/adr/. Computer/Screens, PinchTab, name, Vite PWA, CONTRIBUTING. Harness lives on the host OS: ADR 0005 ([grilling] Where the Harness process lives).

Layer 1 (picked)

Research, Eyes, CONTRIBUTING. Spec: [spec] OpenBot v1.

Layer 1 build (done)

1. [task] Open the app — done
2. [task] Watch a desktop — done

Frontier (Talk / Harness)

1. [research] Where Grok Bot's Harness actually runs — done
2. [research] t3code as OpenBot ACP host — done
3. [grilling] Where the Harness process lives — done, ADR 0005
4. [grilling] Talk voice: short replies and screenshots — HITL next

Then the remaining tracers:

• [task] Talk to one Bot — topology locked; grill Talk voice first
• [task] Two Bots share logins and Sleep — blocked by Talk
• [task] Takeover for 2FA — blocked by Talk

Not yet specified

Cookie encryption at rest. Selkies if Kasm fails. Group chats. Plugins. Routines. ntfy. App Store wrap. Arena. Paid credits.

Parked

Arena / Twitch-plays, hosted proxy on toolbox.md, native app, CopilotKit fork. Laptop CLI driving a remote VPS Screen.

[lukemaj]

Gate 1 checkpoint, 2026-08-29.

The Frontier section in the original map is historical and is no longer executable task order. The current dependency frontier is:

#124 -> #116 -> #80 -> #83 -> #84 -> #85 -> (#70 and #86) -> #87 -> #99

#124 is the sole agent-running implementation. There are currently no ready-for-agent issues. #21 is the only immediate human confirmation ticket. After #124 is live-accepted and integrated, #116 owns reviewed convergence of the familiar-Messenger and isolated-runtime non-main trains before any combined-train implementation or acceptance work.

Do not dispatch work from the old Talk/Harness list above. Current active intent lives in #71, #81, #91, #116, their linked children, and the exact train checkpoints recorded there. This comment updates tracker navigation only and preserves the original map as history.

[lukemaj]

Current executable frontier, 2026-08-30

The previous Gate 1 comment is historical. The current repository and tracker frontier is:

#116 exact-head live/human disposition -> corrected #137/#139 re-anchoring -> #83 live/publication -> #84 live acceptance -> #85 decision -> (#70 and #86) -> #87 -> #99

Current facts:

• [bug] Keep Talk available while persisted Screens recover #124 is live-accepted, merged to the isolated train, preserved in the corrected combined branch, and ready-for-human.
• PR feat: converge familiar and isolated v1 trains #136 is the corrected combined-train candidate at 47923fd005590ffa57bf161fdb32bf6cc50221dd.
• PRs fix: contain selected-Bot Instructions layout #138 and fix: give the phone Chat composer full touch targets #140 are still predecessor-rooted and must remain frozen while feat: converge familiar and isolated v1 trains #136 is open.
• [task] Honest Isolated config and Host config contract #83 is a frozen local deterministic candidate at 22d95a008d52a93f6d9c4d351739466914a1ea11.
• [task] Disconnect Host grants from v1 authorization #84 needs no new code. Exact-[task] Honest Isolated config and Host config contract #83 active/dormant permission acceptance is 79/79; real-provider proof remains unchecked.
• [task] Spawn Grok Build over ACP #34 and the local [task] System Turns for group membership #62 through [task] Pin and drag #66 successor chain remain separate live/publication work, not proof for the corrected PR stack.
• [task] Screenshot Card and attach an image #41 and [task] Give new Sessions the mode-aware OpenBot environment brief #85 still require human product decisions before implementation.
• [grilling] Talk voice: short replies and screenshots #21 is the immediate human inspection contract when the user requests the pause.
• GitHub currently has exactly three open PRs: feat: converge familiar and isolated v1 trains #136, fix: contain selected-Bot Instructions layout #138, and fix: give the phone Chat composer full touch targets #140.

Do not dispatch from the original issue list as linear order. Use the current issue checkpoints and exact branch ancestry. This comment updates navigation only and performs no product or publication mutation.

[lukemaj]

Verified local successor train delta

Fresh worktree, parent, version, remote-head, and remote-tag checks establish this exact clean local chain:

47923fd (#116 v0.15.0) -> be59df29 (#137/#139 v0.15.2) -> 22d95a00 (#83 v0.15.3) -> 8ada14d5 (#34 v0.16.0) -> 6282e976 (#62 v0.17.0) -> 0dffb4d7 (#63 v0.18.0) -> 6401ca89 (#64 v0.19.0) -> 9a4de533 (#66 v0.20.0) -> b85d6b7d (#65 v0.21.0).

Every listed successor worktree is clean. The exact candidate branch names are absent from origin, and tags v0.15.2 through v0.21.0 are absent. These are local-only preserved candidates, not independently mergeable slices.

Deterministic full-suite authority along the frozen local layers is:

• [task] Honest Isolated config and Host config contract #83: 849/849
• [task] Spawn Grok Build over ACP #34: 862/862
• [task] System Turns for group membership #62: 863/863
• [task] Settings sidebar and hide #63: 877/877
• [task] Add, remove, Create Bot from group #64: 883/883
• [task] Pin and drag #66: 898/898
• [task] Screen grid zoom #65: 910/910

Remaining proof and authority are separate:

• [task] Converge familiar Messenger and isolated-runtime trains #116 exact-head live and human disposition are pending.
• corrected [bug] Keep selected-Bot Instructions inside its panel #137/[bug] Give the phone Chat composer full touch targets #139 be59df29 is unpushed and not exact-build live accepted.
• [task] Honest Isolated config and Host config contract #83 and [task] System Turns for group membership #62 through [task] Pin and drag #66 lack exact-build live acceptance.
• [task] Spawn Grok Build over ACP #34 real Grok provider proof requires explicit credit-spending authority.
• no successor push, PR, tag, release, deployment, or installation exists.

PR freeze evidence:

• feat: converge familiar and isolated v1 trains #136 exact-head proof and live boundary: feat: converge familiar and isolated v1 trains #136 (comment)
• fix: contain selected-Bot Instructions layout #138 predecessor-root warning: fix: contain selected-Bot Instructions layout #138 (comment)
• fix: give the phone Chat composer full touch targets #140 stacked-child warning: fix: give the phone Chat composer full touch targets #140 (comment)

The exact #60 stack checkpoint is #60 (comment).

No issue currently carries agent-running. This means all prior writers have released ownership, not that blocked live or human gates have disappeared.