Code

[Reborn] Expose verbatim resp_… (previous_response_id) to tools for external-API conversation continuation (engine v2 #3669 parity)

Define strict subagent cancel and rollback cleanup policy

Reborn compaction retry needs status-only stabilization metadata

[QA] Sandbox docker mount failed on Windows due to Extended-Length Path prefix (too many colons)

Make automation triggered runs clickable in WebUI

PR 18.8: Python E2E cron trigger scenario (blocked on Reborn production profile wiring)

Reborn: regression test — every visible capability must be callable (visible_capabilities ⇔ tool_definitions parity)

Reborn: prompt bundle rebuild waste — ThreadBackedLoopContextPort reconstructed per model call; identity/skill caches cold every time

Reborn: builtin.skill_list unbounded — returns full descriptions for all installed skills; no skill_search exposed to model

Reborn: loop exit reason invisible — LoopFailureKind never traced, only persisted to DB

Reborn: parent loop tool surface is AllowAll — 'interactive_tools' profile_id stamped but ignored; lifecycle/mutation tools exposed in every chat

Consolidate trigger poller test-support handles into TriggerPollerTestHandles struct

IronClaw may fail to start due to stale PID file

`/model` returns display names that cannot be used to switch models with NEAR AI provider

Harden HTTP credential carriers with non-clone types and DTO boundary

[reborn-loop] L11: Architecture + convention hygiene — LoopHostDependencies bundle, debug! demotion, saturating_sub

[reborn-loop] L10: Loop strategies + stop conditions — drain cap, identity bin-pack, JSON reply rejection

[reborn-loop] L9: Compaction + checkpoint — defer-not-error, pre-size check, drop hex encoding

[reborn-loop] L8: Cancellation propagation — interruptible model awaits + polling backoff

[reborn-loop] L7: Budget accuracy — wall-clock, output-token reconcile, cost coverage, CJK tokens

[reborn-loop] L6: Capability pipeline observability + batch rollback + system_inference lifecycle

[reborn-loop] L5: Persistence + event ordering + F-A1 re-investigation

[reborn-loop] L4: Recovery strategy coverage + exit checkpoint fallback + retry surface refresh

[reborn-loop] L2: Prompt safety wiring — required safety_context, SafetyLayer-scanned skills, deprecate bare gateway

[reborn-subagent] C6: Hygiene — readiness accuracy, eviction observability, config knobs, graceful unknowns

[reborn-subagent] C5: Subagent provenance in audit envelope, capability results, and effect kinds

[QA] MCP integration acknowledged but unusable due to driver failure (Qwen3.6-35B-A3B-FP8)

[QA] Authentication modal persists after page refresh and blocks chat (Qwen3.6-35B-A3B-FP8)

[QA] Content field blank validation error blocks message submission (Qwen3.6-35B-A3B-FP8)

[QA] Provider tool calls rejected as InvalidInvocation despite valid capability schema (MiniMax-M2.7)