Warning: Full Suite failed for this source. A Full Suite failure report ends these notes.
Merged commits since preview source 3b3beb5:
- fix(plan-approval): a message the agent names counts, an old recovery question waits for a reply after it, and what to change is what the person said (#2951)
- fix(reverse-engineering): the gate over many repositories checks their files in one pass and names what is missing (#2940)
- fix(plan-approval): a Code Generation completion that comes too early gets the engine's one step, not a guard refusal first (#2946)
- fix(guards): while a plan waits, the doctor, status and next run after a cd to the project (#2699)
- fix(settings): a setting typed while the questions are written goes on to them, and finished work names no place to pick up (#2945)
- fix(questions): a question the agent logs after the person replied records on that reply, with nothing read back to them (#2938)
- fix(compose): approving a trimmed plan mid-work carries on into the next step (#2944)
- fix(windows): a dev server the AI left running is stopped by the AI, so the person is no longer asked to close a program they never opened (#2929)
- test(reading): a pin and a text-blind fuzz fail the day the engine reads the person's words again (#2939)
- feat(intent): add a repo to a piece of work at any stage, cloning it when it is not here yet (#2934)
- fix(locks): a hook killed while it takes a lock no longer stalls every later Write and doctor for 5 minutes (#2937)
- fix(units): a skip of the stage a Unit's walk starts on is carried out, for that Unit or every Unit (#2928)
- fix(questions): under "Continue automatically" a question the stage asked stays open until the person answers it (#2933)
- test(gate-fuzz): the engine's outcome at every gate and question stays the same whatever words the person types (#2935)
- fix(summary): the summary confirmation counts on a reply after its own question, or on the message the agent names (#2931)
- fix(recompose): under "Continue automatically" the person's own skip or add changes the plan (#2932)
- test(loops): an agent's fix at any point the person is asked leads back to them, never into a refusal loop (#2930)
- fix(decisions): a decision the agent records is checked for the person's reply alone, not for what its text says about who decided (#2927)
- fix(questions): at the engine's own questions the agent records the choice it read from the person's reply, in any words (#2926)
- fix(plan-approval): a check the evaluator never reached reads as not evaluated (#2878)
Source commit: 085d534
Nightly test report
Full Suite failed for 085d534c4de2 in run 40025954718. The preview build does not wait for these tests.
Failed legs: native_terminal (failure), native_reconcile (failure).
Failing tests
No failing test files were recorded in the downloaded evidence. The failed jobs below may have stopped before or outside their tests.
Failed jobs
- Linux / native-reconcile (1): Linux / native-reconcile
- Linux / result (1): Linux / result
- macOS / native-terminal (1): macOS / native-terminal bun
Full logs are in the run 40025954718 full-suite-* artifacts.