Repository navigation
1.42: credits at the wall, and relay cloud - #1
Merged
Merged
Conversation
…, and relay cloud hands the work to one This account has usage credits switched off at the org level (cachedExtraUsageDisabledReason "org_level_disabled"), so at 100 per cent of the 5-hour window a local session stops dead and the report said so and nothing else. The CLI itself offers a second way at that same wall: a dollar credit for cloud sessions, on top of the plan (tengu_swift_lynx, read from ~/.claude.json in a cloud container on 2026-09-28), and a cloud session on it was seen still working with its own record saying five_hour "rejected". lowpri.cloudCredit reads the offer: the amount is parsed from the server's own copy, never written here, and an absent, malformed or amount-less flag is no offer. The report gets a Cloud credit line and, at the wall with usage credits off or their spend limit reached, one line naming it as the way on now; the resume-after line stays. The brief says it once a session near the wall, in both styles, and not when usage credits are on or when it is already running in a cloud session. Inside a cloud container (detected from CLAUDE_CODE_CONTAINER_ID, injectable) there is no usage snapshot, so the report no longer says "Run /usage once", which cannot work there; it says what kind of session this is and what it has cost so far at list prices, subagents included, against the offered amount. relay cloud builds the prompt the wake would deliver, opened for a cloud session, inline because a cloud session cannot read a file here, and prints the claude --cloud command quoted for sh or PowerShell. It warns about uncommitted changes, unpushed commits and a branch with no upstream, because the cloud clones from GitHub. --go launches it detached with its output in a run log, and is guarded like the scheduler: a dry run under node --test, with USAGE_LIMITS_CLOUD=dry-run or a redirected home, and every test launch goes to a fake spawner. test/cloud-credit.test.js (16) and test/relay-cloud.test.js (13); all 29 fail against the scripts before this change. Full suite 970/972, 2 skipped (Task Scheduler is Windows only; vscode/lib not built), 0 failed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CPrhYxAHgTrw4QZVtWaHV8
package.json, both plugin manifests and the VS Code package say 1.42.0, and vscode/lib is rebuilt from the scripts at that version - 32 scripts. 971/972, the one skip being Task Scheduler, which is Windows only; no usage-limits temp folder left behind. Minor, because relay cloud is a new command. A correction to the previous commit's body: the new files hold 15 and 13 tests, 28 in all, not 16 and 13. All 28 fail when run against the scripts at 1.41.1 (checked from a git archive of 7dadccc), and the suite there was 970/972 with 2 skipped, as it said. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CPrhYxAHgTrw4QZVtWaHV8
…d stops sending a batch shim to PowerShell Three things a review of 1.42.0 found in the cloud hand-off. The hand-off told the cloud session it was a fresh clone with only what was pushed, and the warnings told the user to push the branch, but nothing told the cloud session which branch that was. Which branch a cloud clone starts on is not ours to choose, so work pushed to a feature branch could read as missing, and the prompt says to stop and say so when something is missing. The branch is now named in the cloud opening (not for a detached HEAD, and never in the wake's own prompt, which resumes in place). PowerShell also closes a single-quoted string at a curly single quote (U+2018-U+201B), and psQuote only doubled the straight one, so a continuation with a typographic apostrophe in it broke the printed command on Windows. They are doubled now, which is also right for the scheduler's uses of psQuote. The --go refusal for a .cmd shim said to paste the command into PowerShell instead, but PowerShell runs a .cmd through cmd.exe too, so that cut the hand-off at the same line break. It now points at the native claude.exe. The new assertions fail without the change (3 of 14 in relay-cloud). Suite 972/973, the skip being Task Scheduler, which is Windows only. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CPrhYxAHgTrw4QZVtWaHV8
package.json, both plugin manifests and the VS Code package say 1.42.1, and vscode/lib is rebuilt from the scripts at that version - 32 scripts. 972/973, the one skip being Task Scheduler, which is Windows only; no usage-limits temp folder left behind. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CPrhYxAHgTrw4QZVtWaHV8
…fore a batch shim cuts the hand-off defer.md said the wake "starts a fresh session in the original directory and hands it the saved plan", but wake.js ran claude --resume <id>: the same conversation. A long session is the one that runs out, and resuming carries all of it into the next window. relay fresh on|off (off by default, so nothing changes unless it is set) picks which conversation mode resume continues in. On, the wake starts claude in the project with the whole hand-off as its first prompt, inline, Remote Control on as "usage-limits relay <project> (new session)", with the relay's permission mode and model, and no transcript needed - so a deferral made where no session id is known now has something that works. The prompt opens by telling the new session it has none of the conversation (and, for a deferral, that it is the deferred time rather than a reset). relay show off makes it a headless claude -p with the hand-off on stdin. ONE TERMINAL is unchanged: an old session open and working again makes the wake stand down; open but quiet for three minutes, the wake opens its one window. Under mode notify the note names the hand-off file to give a new session. Codex resumes its thread either way. On Windows the launcher is a .cmd and cmd.exe cannot carry a line break, so the inline hand-off goes through node (wake.js --exec-spec) to the native claude.exe; with only the npm claude.cmd shim, or a hand-off too long for one argument, the first prompt points at the hand-off file instead, and the relay log says why. UNVERIFIED on a real Windows console: the argument passing is tested through node here, not a live claude.exe window. defer's one-line reply now says what fires at the time (a notification, the same conversation, or a new session), and marks its own record: it passed the sessionId function to armedFor, which matched nothing, so with another relay armed first the deferred label went on that one. relay cloud on Windows now warns, before anything is pasted, when the claude on PATH is a .cmd/.bat shim: PowerShell runs it through cmd.exe, which cuts the hand-off at its first line break. It names a native claude.exe when one is found, which --go already uses. defer.md, relay.md, SKILL.md and the README say what happens in each mode. test/relay-fresh.test.js (13) and one relay-cloud test; all 14 fail against the scripts before this change. Suite 986/987, the skip being Task Scheduler, which is Windows only. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CPrhYxAHgTrw4QZVtWaHV8
package.json, both plugin manifests and the VS Code package say 1.43.0, and vscode/lib is rebuilt from the scripts at that version - 32 scripts. 986/987, the one skip being Task Scheduler, which is Windows only; no usage-limits temp folder left behind. Minor, because relay fresh is a new setting. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CPrhYxAHgTrw4QZVtWaHV8
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.
This PR plans against credits instead of only the 100% usage limit. It was tested in the real scenario: a cloud session running on the $250 cloud-session credit while the 5-hour window read "rejected".
What this account actually has (verified in
~/.claude.json):org_level_disabled, so a local session stops dead at the wall.tengu_swift_lynxflag.Changes:
Cloud creditline. At the wall, with usage credits off, the report and brief name "carry on now in a cloud session" instead of only "resume after HH:MM".relay cloudprints theclaude --cloud "<continuation>"command, and--golaunches it. It warns about a dirty tree or unpushed commits, because the cloud clones from GitHub.Tests: 972 tests: 971 pass, 1 skipped (Task Scheduler is Windows-only). No test launches a real cloud session.
Not verified:
claude --cloudlaunch. That was skipped on purpose, because it would spend credit.Coming to this branch next: a relay mode that starts a genuinely fresh session. Today's wake runs
claude --resume <same id>althoughdefer.mdsays "fresh session".🤖 Generated with Claude Code
https://claude.ai/code/session_01CPrhYxAHgTrw4QZVtWaHV8
Generated by Claude Code
https://claude.ai/code/session_01WtLiUzvgkpgL2BP6qJiQP5