You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
- Fix the 'retry'.includes('try') label-collision class: normalizeStepName
(agent-generics + the operative uapi copy) now tests 'retry' before 'try';
same fix in the PTRR-snapshot step inference and the PathPill step-icon
rules; the header strip normalizes its step chips. A Retry step rendering
as 'Try' after Refine presented as a PTRR sequencing bug — the engine order
plan→try→refine→retry is checker-pinned canon (protocol V38_PTRR_STEP_IDS)
and is unchanged. Retry rows now label, tooltip (STEP_SPECIFICS.retry was
unreachable), icon, gerund ('Retrying'), and selector-highlight as Retry.
- Normalize the PrepareConciseContext display label to 'Prepare Context':
FAILSAFE_PILL_NAMES entry, tooltip title override + generic failsafe copy,
and the PTRR-snapshot raw machine-id leak now formats through formatMeta.
Machine ids (prepare_concise_context et al.) unchanged.
- Add 'Copy terse logs' beside 'Copy raw logs': events compact to
timestamp/type/namespace/key/call-chain/usage/message/error rows (error and
message fields keep a 2000-char budget; other strings 200-char previews);
the outputDetails duplication is omitted. A much smaller payload that keeps
ordering, hierarchy, and failure forensics.
- writeRawLLMIO claims its monotonic sequence number before any await so
fire-and-forget concurrent writes cannot swap on-disk file numbers (the
judge-request-before-reason-response interleavings in run sidecars were
logging artifacts, not engine concurrency).
Verified: uapi jest 155 suites / 597 tests green; agent-generics 19 suites /
99 tests green; uapi tsc --noEmit clean; eslint clean on touched files.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
'A live, source-safe stream of the SynthesizeAssetPacks pipeline actually running — every phase, agent, step, and LLM/tool call, with prompt and response content withheld by law.',
148
148
detail:
149
-
'Rows appear only for completed LLM calls and tool uses, each carrying its full Phase→Agent→Step→Failsafe→Thinkings hierarchy; the processing indicator underneath shows what is currently running and how long since the last update, so a genuine stall is visible instead of an unexplained gap. Use "Copy raw logs" to grab the full run for support/debugging.',
149
+
'Rows appear only for completed LLM calls and tool uses, each carrying its full Phase→Agent→Step→Failsafe→Thinkings hierarchy; the processing indicator underneath shows what is currently running and how long since the last update, so a genuine stall is visible instead of an unexplained gap. Use "Copy raw logs" to grab the full run for support/debugging, or "Copy terse logs" for a compact distillation (hierarchy, ordering, usage, and errors — long bodies truncated).',
150
150
points: [
151
151
'Watch which phase/agent is running in real time',
152
152
'The processing indicator flags a likely stall after ~90s of silence',
153
153
'Copy raw logs exports the full source-safe run for debugging',
0 commit comments