RunningTab introduces an environment-side task record and reports consistent gains across three benchmarks and three LLMs.
RunningTab keeps workspace-agent obligations outside the context window
2026-10-08Frontier watchWATCH
RunningTab adds an environment-side record to direct workspace interaction. It tracks task requirements, excerpts with provenance and listed-but-unopened files, then blocks finishing while obligations remain unresolved. The authors report consistent gains over plain workspace interaction and model-maintained records across three benchmarks and three LLMs.
Availability: paper only; no public code or evaluation harness found
What this means for you
For teams building long-running file agents, this suggests that completeness state belongs in the runtime, not only in model context. Without public code or a harness, the design is actionable as an architecture signal but not yet ready for a team POC.
FRONTIER WATCH
Public artifacts and validated results are different.
The recorded fact is the public work and its artifacts. Author-reported outcomes do not establish independent reproduction. Next checks include usable code, task conditions and third-party validation.
Evidence & related material
Technical details & scores
| Mechanism | environment-side requirements, provenance excerpts, unopened-file candidates and a finish check |
|---|---|
| Reported Evaluation | three benchmarks and three LLMs; consistent gains over plain direct workspace interaction and model-side records |
| Availability | paper only; no public code or evaluation harness found |
| Evidence Limit | author-reported preprint |
Technical impact 3
Engineering value 4
Adoption signal 1
Maturity 1
Verification cost 4
Risk 2