fix(mothership): explain withheld code and workflow results and bound workflow block logs - #8459

Merged

waleedlatif1 merged 14 commits into

stagingfrom

fix/code-execution-provenance

Sep 30, 2026

Commits on Sep 30, 2026

fix(mothership): keep a tainted mount's verdict across the run_code crossing
waleedlatif1
committed
b5ed701
improvement(mothership): tell the model why a tool result was withheld
waleedlatif1
committed
68f42a6
improvement(mothership): log the size of a withheld tool result
waleedlatif1
committed
01ff74b
fix(mothership): bound run_workflow block-log outputs before the secret projection
waleedlatif1
committed
805805a
fix(mothership): bound browser-run workflow logs and keep the pointer length-blind
waleedlatif1
committed
b4fc3c4
fix(mothership): resolve a browser run's select before the secret projection
waleedlatif1
committed
b44e3c1
test(providers): expect the withheld reason on provider tool results
waleedlatif1
committed
ec785db
fix(mothership): bound lifted outputs, length-blind input markers, and a walk-based measure
waleedlatif1
committed
42e012e
fix(mothership): keep no raw prefix in a truncated block input
waleedlatif1
committed
4b143ce
fix(mothership): bound run outputs only when the projection walks them against a secret
waleedlatif1
committed
0cb1559
fix(mothership): measure a run's whole model-facing result and bound its final output
waleedlatif1
committed
118a58b
fix(mothership): leave an unencodable block output for the projection to refuse
waleedlatif1
committed
e4ba8a3
fix(mothership): leave a result that fits the caps, and every result without a secret, as staging returns it
waleedlatif1
committed
2c0b369
test(mothership): pin the unencodable-output guard and the error in the whole-result measure
waleedlatif1
committed
e25849b

Navigation Menu

fix(mothership): explain withheld code and workflow results and bound workflow block logs - #8459

fix(mothership): explain withheld code and workflow results and bound workflow block logs#8459
waleedlatif1 merged 14 commits into
stagingfrom
fix/code-execution-provenance

Commits on Sep 30, 2026

fix(mothership): keep a tainted mount's verdict across the run_code crossing

improvement(mothership): tell the model why a tool result was withheld

improvement(mothership): log the size of a withheld tool result

fix(mothership): bound run_workflow block-log outputs before the secret projection

fix(mothership): bound browser-run workflow logs and keep the pointer length-blind

fix(mothership): resolve a browser run's select before the secret projection

test(providers): expect the withheld reason on provider tool results

fix(mothership): bound lifted outputs, length-blind input markers, and a walk-based measure

fix(mothership): keep no raw prefix in a truncated block input

fix(mothership): bound run outputs only when the projection walks them against a secret

fix(mothership): measure a run's whole model-facing result and bound its final output

fix(mothership): leave an unencodable block output for the projection to refuse

fix(mothership): leave a result that fits the caps, and every result without a secret, as staging returns it

test(mothership): pin the unencodable-output guard and the error in the whole-result measure