Skip to content

v0.5.2

Choose a tag to compare

@github-actions github-actions released this 09 Nov 04:21
· 32 commits to main since this release

Added

  • Relocatable profiles root: LISA_PROFILES_DIR moves the profiles root
    (<profiles-dir>/<name> layout unchanged), so profiles can live in a
    synced folder while the packs venv stays machine-local. A relocated
    profile shares the base data dir's packs venv. Kernel fail-closed
    hardening: missing lifecycle targets and tool errors under an abort
    pre_turn_failure_policy fail the turn instead of proceeding.
  • Memory pack evals (tests/evals/test_pack_validity.py, eval marker) and
    an offline scripted e2e scenario (tests/e2e/dedup_recall.scenario.yml,
    registered in e2etest/packs.yml) that proves the store-then-recall flow
    end to end with surface-scoped records.

Changed

  • Memory capture is now a write-time upsert (breaking). The memory pack's
    synthesize_and_store_learning and mine_thread_segment operations search
    the scoped store for similar existing records before distilling and persist
    a per-fact create/update/skip decision: updates write a superseding record
    (supersedes_id) and retire the scoped active original in one transaction;
    duplicates are skipped. Captures are stored as memory_kind="semantic" —
    distilled facts, not raw
    conversation events. This replaces the batch
    consolidate-episodic-pressure pipeline (the removed
    consolidate_semantic operation), aligning the write path with the
    extract-then-dedup design current agent-memory systems converge on.
  • context.memory.consolidation (MemoryConsolidationTuning) is removed
    (breaking) and replaced by context.memory.mining
    (MemoryMiningTuning): enabled, trigger, preferred_model_id,
    mine_operation, max_segments_per_slice. Thread mining keeps running on
    the slice provider under the new tuning.
  • dispute_memory performs a real supersession: the correction carries the
    supersedes_id column link, inherits the disputed record's kind and scope,
    and the original record is retired so it drops out of the active recall
    pool (still queryable with status=None for audit). Previously the
    original stayed active at full salience forever.
  • Memory recall uses one shared tokenizer (sdk.memory.lexical) across the
    store's recall pools and the pack's evidence ranker, and the operation asks
    the backend for candidate headroom above the final evidence limit so the
    relevance-first re-ranker sees the full pool before truncating.
  • The memory-mining slice is fixed for production wiring: the slice-provider
    view now carries the durable data root, active user, and model seams, and
    the mining slice builds a real OperationExecutionContext instead of a
    bare params dict. Previously the factory left memory_data_dir=None and
    the user slug defaulted, so mining and consolidation silently never ran.
  • The mine.marker sidecar is positional: a marker id missing from the
    segment index (rebuilt store) re-mines from the start instead of silently
    stopping forever — non-destructive because write-time dedup skips records
    that already exist.

Fixed

  • Thread mining leaves a segment pending after an operation exception or
    memory-write failure, stopping the pass before a later segment can advance
    the marker past it. A partial write is retried using write-time dedup.
  • Slice-provider content reads report an uninitialized content registry with
    setup guidance instead of an attribute error.
  • Memory capture and mining log malformed dedup-search responses before
    proceeding without candidates.
  • A failed supersession transaction rolls back its successor and retirement;
    failed dispute writes report an error status instead of a successful result.
  • Supersession rejects a missing target ID that would make the successor
    reference and retire itself.
  • Dispute and mining report the retirement performed by each superseding
    write rather than assuming a predecessor was still active. Providers that
    do not report the outcome leave it explicitly unknown.
  • File-backed approval and recovery stores fail closed on corruption. An
    unreadable, malformed, or non-object store file, or an invalid record,
    raises ApprovalStoreCorruptError or RecoveryStoreCorruptError instead of
    loading as empty, so the next mutation can no longer overwrite pending
    approvals or open recovery records.

Removed

  • MemoryRecallContext (unused) and the vestigial
    used_semantic_search / auto_deep_used / uncertainty_note /
    semantic_score / recency_score recall fields; the dead
    _store_mined_facts helper; the deep recall parameter (the deep-synthesis
    flow it described never existed).

Migration

  • Replace context.memory.consolidation with context.memory.mining in
    session configuration. Remove calls to consolidate_semantic and the
    deep recall parameter. See
    memory mining configuration.
  • External memory.add providers must atomically retire the scoped active
    predecessor when supersedes_id is supplied; capture, mining, and dispute
    no longer issue a separate memory.retire call. Superseding writes now
    return a JSON metadata line containing memory_id and boolean
    supersession_retired before the built-in server's existing text receipt.
    External providers should return the same metadata so retirement can be
    confirmed; see memory architecture.
  • dispute_memory's retired result can be null when the provider does not
    confirm the outcome. Mining's retired_ids contains only confirmed
    retirements; not_retired_by_write_ids and retirement_unknown_ids report
    the other outcomes, and supersession_outcomes associates each outcome
    with its predecessor and successor.
  • Capture results no longer expose retired_ids; inspect the stored
    successors' supersedes_id links when the prior IDs are needed.
  • A corrupt approval or recovery store file is no longer silently reset.
    Repair or remove the file explicitly when
    ApprovalStoreCorruptError or RecoveryStoreCorruptError is reported;
    a missing file is still treated as empty.

Full Changelog: v0.5.1...v0.5.2