Dashboard › opencode › Distillation
4232d8d3-d2a1-44b2-a3f2-66601f5c8059["lore_tm_v1_Si55G0Dm4uUxrmXjaTDp1xSY0ns1IvcU5pDYjW8Nd24","lore_tm_v1_i6p2Kru_S4Q6SI-la_N_YX96BL5cDG52VgF_miti52s","lore_tm_v1_51r4eDnrrviI57fNyqe_YMZPFx_FrQB2QlJvjMP7Yq8","lore_tm_v1_1KyqEz3aKIzN3cCPiKT1mOtind6xjtFDikaB999hsXM","lore_tm_v1_ayhSVXkz2gOzVv_x88hr-QxAJvmAjpzF2wG4ohIYp4Y","lore_tm_v1_2Vs014RuQbyTSLI4O09wPr5M-RAlFLRr4mvZlNvRXsw"]
Date: Sep 17, 2026
packages/core/test/session-runner.test.ts line 752.packages/core/test/session-runner.test.ts line 778.packages/core/test/session-runner.test.ts line 805.packages/core/test/session-runner.test.ts line 845.packages/core/test/session-runner.test.ts line 1975.packages/core/test/session-runner.test.ts also covers recovery and execution-intent behavior: redriving only the synthetic recovery continuation (line 619), clearing durable execution intent when ordinary input is promoted (line 688), redriving an ordinary prompt after a lost advisory wake (line 715), and interrupting a source Location runner after a Session moves (line 1034).packages/core/test/session-runner.test.ts covers sampled configuration stability: keeping the sampled agent when selection changes during observation (line 1225) and keeping the sampled model when selection changes during model resolution (line 1257).packages/core/test/session-runner.test.ts covers failure-boundary durability: never committing a Step end after provider-stream failure (line 2033), keeping a running assistant recoverable when Step failure publication defects (line 2055), never recording assistant start state when Step start publication defects (line 2091), durably reconciling a local tool when success publication defects (line 2128), keeping a provider defect after Step finish is unknown (line 2167), and keeping provider failure unknown until unresolved tools are durable (line 2190).packages/core/test/session-runner.test.ts covers local/hosted tool sequencing: continuing with reloaded history after durably settling one local tool call (line 1917), recovering continuation committed with a settled tool step (line 2238), reloading a model switch before a tool-driven continuation (line 2303), replaying durable provider-executed tool results inline (line 2405), eagerly starting recorded local tools while awaiting settlement before continuation (line 2460), settling repeated provider-local tool-call IDs against their owning assistant messages (line 2521), durably failing abandoned local tools before continuing (line 3035), and durably failing abandoned hosted tools before continuing inline (line 3095).packages/core/test/session-runner.test.ts covers concurrency and input ordering: joining concurrent resume calls into one active provider run (line 2609), steering an active provider turn with newly recorded prompts (line 2648), promoting queued input after continuation ends (line 2691), preserving queued and steering inputs after interruption (lines 2738 and 2781), promoting queued inputs individually in FIFO order (line 2824), promoting steers before the next queued input (line 2901), coalescing multiple active steering prompts into one continuation turn (line 2966), and running steering input accepted while the active provider turn fails (line 3006).packages/core/test/session-execution-local.test.ts verifies claim/lease fencing: successful drains clear claims (line 45); acquisition snapshots the latest assistant sequence (line 54); scope closure preserves the claim (line 86); explicit interruption clears it (line 102); a live lease prevents a second owner (line 116); expired leases are never overwritten (line 139); an owner never re-enters its own uncertain live lease (line 166); heartbeat expiry is renewed from the current clock (line 193); an expired claim is never revived by heartbeat (line 216); cancellation settles before a pending successor starts (line 234); runner settlement cannot overwrite cancellation (line 278); and cancellation fails if cleanup loses execution ownership (line 313).packages/core/test/session-projector.test.ts verifies recovery fencing against live or superseded ownership: never recovering a live execution lease (line 304), never recovering an execution superseded by a newer assistant (line 342), never promoting a recovery superseded before scanner validation (line 1699), never recovering an expired cancellation claim (line 2018), clearing an expired durably cancelled claim (line 2069), rolling back promotion and Context Epoch mutation after ownership loss (line 2543), never advancing the Context Epoch after ownership changes (line 2603), rolling back claim renewal when Context Epoch advance fails (line 2696), and fencing a paused owner after expiry, reconciliation, and takeover (line 2787).packages/core/test/session-projector.test.ts covers deterministic and bounded recovery scheduling: reconciling one abandoned execution into one continuation atomically (line 68), recovering each abandoned execution once across repeated startup scans (line 258), redriving durable recovery wakes after admission (line 555), quarantining malformed or poison recovery rows without blocking valid work (lines 606 and 660), fairly redriving recovery and ordinary work beyond a full bounded batch (line 725), bounding ordinary redrive writes to one scheduler row per Session (line 951), never waking work settled between selection and redrive CAS (line 1011), rotating maximum wake-attempt counters before wakes (line 2423), and bounding each deterministic recovery scan (line 2498).packages/core/test/session-projector.test.ts validates recovery identity isolation: quarantining a continuation belonging to another Session (line 1094), rejecting a recovered assistant from another Session (line 1436), rejecting a recovery execution identity owned by another Session (line 1483), quarantining a claim blocked by a conflicting recovery identity (line 1535), quarantining a recovery wake behind a newer assistant (line 1584), never quarantining an exact recovery conflict after ownership transfer (line 1763), and never promoting a quarantined recovery continuation as an ordinary steer (line 1910).packages/core/test/session-prompt.test.ts verifies prompt admission and wake semantics: durable admission before transcript promotion (line 181), exact retries return the original message (line 290), exact retries recover committed messages by waking execution (line 310), retries of already promoted prompts never wake execution (line 330), durable execution intent is recorded before advisory wake (line 374), historical admission without intent replays as admit-only (line 408), an exact retry can turn admit-only input into durable execution intent (line 440), conflicting execution retries never arm durable wake (line 514), and pending inbox input can be reprojected without scheduling execution (line 642).packages/core/test/session-prompt.test.ts verifies prompt identity and concurrency: reuse of one ID with a different prompt is rejected (line 489), reuse with a different delivery mode is rejected (line 556), concurrent exact retries return one recorded message (line 581), concurrent promotion attempts promote one message once (line 601), steers are promoted only through the captured inbox cutoff (line 625), globally unique message IDs cannot be reused across Sessions (line 710), and prompt IDs already used by visible Session history are rejected (line 740).packages/core/test/session-prompt.test.ts establishes resume defaults: prompt recording starts execution by default (line 761), resume: true explicitly starts execution (line 775), and resume: false records the prompt without starting execution (line 793).