@steipete @ratulsarna
I found several problems in the old Manus PR #476 provider path, and I already implemented a clean main-compatible fix locally.
Problem
PR #476 had a few issues that make it hard to land safely on current main:
- token source semantics were leaky
- manual / off-mode boundaries were not strict enough
- invalid manual input could fall through to other sources instead of failing fast
- Manus usage was forced into generic credits / extra-usage UI that does not match the actual Manus quota model
- the branch has drifted too far from current
main to be a safe merge candidate
What I fixed locally
I rebuilt the Manus provider cleanly on top of current main:
- provider descriptor + implementation + settings wiring
- bounded auth resolution:
off stays off
manual does not silently fall back
- browser/cache/env fallback is explicit and ordered
- Manus-specific usage mapping:
- balance
- monthly credits
- daily refresh
- official Manus brand asset integration
- dedicated Manus tests for cookie parsing, fallback behavior, cache behavior, and sparse live payload parsing
Validation
swift test --filter Manus passed locally
pnpm check passed
./Scripts/compile_and_run.sh passed
- verified against a real live Manus session and the live Manus credits API
I found the problem, and I already fixed it locally as a clean replacement for PR #476 rather than trying to revive the drifted branch.
@steipete @ratulsarna
I found several problems in the old Manus PR #476 provider path, and I already implemented a clean main-compatible fix locally.
Problem
PR #476 had a few issues that make it hard to land safely on current
main:mainto be a safe merge candidateWhat I fixed locally
I rebuilt the Manus provider cleanly on top of current
main:offstays offmanualdoes not silently fall backValidation
swift test --filter Manuspassed locallypnpm checkpassed./Scripts/compile_and_run.shpassedI found the problem, and I already fixed it locally as a clean replacement for PR #476 rather than trying to revive the drifted branch.