Session 2026-08-25T20-25-33-180Z - #1706
Closed
suleimansh wants to merge 5 commits into
Closed
Conversation
…session list (#1332) The extension used to open one pinned tab per cloud session, capped at the three newest. It now keeps one pinned "The Framework Driver" tab that reads claude.ai's session list (each row links to its session and carries a text status label), reports those statuses to the daemon, visits only the sessions awaiting input, unread, or holding a queued answer — navigating inside the app, with one page load a minute to refresh the list — types the answers, and creates the sessions the daemon asks for. A full-page overlay names the tab and holds the cycle log behind "Show debug logs"; closing the tab pauses the bridge until the options page reopens it. Daemon: the session list loses its cap and gains answerQueued; POST /_bridge/statuses stores what the list said per session; a session the list shows awaiting input is waiting on a human, so its row says so even when the question was asked in prose. Lockstep bump to 0.12.0. Measured on the live page: the list does not refresh on in-app navigation, and a visit clears neither "Awaiting input" nor "Unread response", so visits are planned on change, age or a queued answer (driver-plan.js, shared by the worker and the harness).
…ch (#1332) Review fixes on the Driver-tab PR. The tab id and the pause live in session storage, so no onStartup race and no stale id hijacking a tab; a Driver tab moved off claude.ai is forgotten rather than dragged back. A reload that fails, or a page torn down mid-drive, fails the cycle instead of throwing past the accounting or re-sending a drive that is not idempotent. At most four visits per cycle, answers first, and a visit cut or failed keeps its change pending. The page refuses a second drive while one runs. The requested session is created before the visits. The list is waited for before it is read and a page with no rows is named; a status word beats a pull-request label; a link elsewhere on the page does not shadow a list row; a visit waits for the new session's rows and a send counts by a new row, not a higher count. An answer the visit never reached stays queued instead of failing. An off or paused Driver still pays its acks and fails a session request at once. Closing the window is not closing the tab. A list status older than the session window no longer marks a run waiting, and sessions holding a queued answer are served whatever the window says. SPEC files reworded to what the code does.
…'s memory (#1332) Found dogfooding the Driver tab. A cloud session's question had been answered two hours earlier; restarting the daemon emptied the store that remembered the delivery, the Driver's next visit re-reported the block still on the page, the dashboard asked again, and the second answer was typed into a session that had already finished — the cloud agent received two different answers to one question and rewrote its branch. The Driver made this systematic: it revisits unread and parked sessions every five minutes, so every daemon restart re-asked every recently answered question. The page half now places the last block in its transcript row — through any shadow root it renders behind — and drops it when a human turn sits at a later position: the user answered, the session moved on. A block in no row is measured from the latest assistant turn; a page marking no turns cannot be measured and its block stands. The panel says "no (already answered)" rather than "no". Three harness cases: answered in the light DOM, answered behind a shadow root, and asked again after an answer — the first two break-checked against the rule removed.
Dogfood of the Driver tab fix (PR 1703, create path). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TU8F5wpEheeZoGGaGENnVU
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Dogfood of the Driver tab fix (PR 1703, create path). Step 1: before touching any file, stop and ask me through the choice gate which marker file to create:
dogfood-1703-create-a.mdordogfood-1703-create-b.md. Step 2: after my answer, create only that file at the repo root containing the single linedriver tab dogfood, commit it, push, and finish. Do nothing else.Opened from The Framework session
2026-08-25T20-25-33-180Z.