v0.2.5

Current release

These changes ship in the Phantasm v0.2.5 release.

Fixed

  • Old v1 stores that were created before records.quality_json and suggestions.quality_json now repair those columns automatically with {} defaults.
  • compile, search, health, and mutating request paths no longer fail with no such column: quality_json against older bootstrapped projects.

Notes

  • Users do not need to rerun phantasm bootstrap or manually edit SQLite stores. Upgrading and rerunning the previously failing command applies the repair.

v0.2.4

These changes ship in the Phantasm v0.2.4 release.

Added

  • Local interactive dashboard through phantasm dashboard, with loopback binding, per-session token auth, embedded offline assets, charts, graph exploration, and review actions.
  • phantasm agents-plugin for installing the bundled harness-agnostic phantasm-memory plugin into supported local agent harnesses.
  • phantasm init as an exact alias for phantasm bootstrap.

Changed

  • bootstrap --agent-guidance now installs the bundled agent plugin by default; use --no-agent-plugin to opt out.
  • Repository and internal docs now reflect local-only git workflows and the completed beads backlog.
  • Dashboard attention links now navigate correctly after dynamic rendering.

v0.2.3

These changes ship in the Phantasm v0.2.3 release.

Added

  • First-class quality metadata for record confidence, source, verification, expiration, revalidation, and authorship, including search and compile filters.
  • Read-only stale_review for freshness and changed-file impact, plus duplicate_review for semantic overlaps and consolidation guidance.
  • memory_export for canonical importable JSON and review-only Markdown, plus additive memory_import for reviewed canonical JSON.

Changed

  • Updated API discovery, CLI help, managed guidance, public docs, site pages, and release notices for the v0.2.3 runtime surface.
  • Sensitive memory exports now require explicit inclusion and an export-capable client profile.

v0.2.2

These changes ship in the Phantasm v0.2.2 release.

Added

  • Added human-friendly memory CLI commands for routine workflows: phantasm list, phantasm show <id>, phantasm search "query", phantasm add --kind <kind> --subject <key>, phantasm stats, and phantasm status.
  • Added a default inspect summary when called without object IDs, including record counts, live/conflict counts, lifecycle state counts, scope counts, review item count, timestamp bounds, and compact live and conflict record lists.
  • Added opt-in compile.params.explain=true diagnostics so callers can see why compile candidates were selected or omitted, including candidate rank, selection rank, token estimates, focus-subject influence, lifecycle state, and token-budget omission reasons.
  • Added ingest.params.review_required=true so trusted callers can intentionally create suggestions and review items instead of direct authoritative memory for automatic or uncertain writes.
  • Added write_mode to ingest dry-run responses so callers can confirm whether a request would write authoritative memory or queue a suggestion.
  • Added a read-only templates operation with built-in templates for project identity, architectural decision, operational constraint, known issue, release procedure, environment setup, and deferred task records.
  • Added scripts/scale_test.sh, a configurable scale harness covering sequential ingest, concurrent writers, search, compile, snapshot restore, and corrupt-snapshot recovery checks.

Changed

  • Updated describe, built-in help, managed agent guidance, public command docs, and the public site to document review_required, compile.explain, templates, the default inspect summary, and human-friendly memory commands.
  • Updated internal operator documentation with quick and default scale test commands and guidance on when to run scale/recovery checks.

v0.2.1

These changes shipped in the Phantasm v0.2.1 release.

Added

  • Added operation-specific CLI help for handle-request so callers can print a single operation schema and runnable example without sending a runtime request first.
  • Added params.dry_run preview support for mutating operations. This lets callers inspect no-write outcomes such as ingest would_create/would_update results before risking collisions or writes to the wrong scope.
  • Added payload fidelity metadata on write responses so callers can compare request bytes and hashes with the stored payload.
  • Added mojibake detection warnings on write responses so UTF-8 transport corruption is surfaced immediately instead of being silently persisted.

Changed

  • Updated unsupported operation validation errors to list all valid operation names and point callers to describe target=all or target=* for the full operation catalog.
  • Updated required-field validation errors to be operation aware. They now include the missing or invalid params.<field>, the expected type, the required param set for that operation, the field description, a runnable example request, and a describe target=<operation> recovery hint.
  • Updated health to expose grouped live_conflict_subjects alongside live_conflict_count so agents can identify affected record_kind/subject_key pairs without scanning audit history.
  • Updated live_conflict lifecycle effect payloads to include the affected record_id, record_kind, and subject_key for faster recovery through the audit surface.
  • Updated built-in help, operation schemas, and public/internal command references to document dry_run previews and the newer operation-specific help surface.

Fixed

  • Fixed snapshot bundle export to capture SQLite WAL-backed state atomically instead of risking inconsistent snapshot contents.
  • Fixed raw-evidence attachment persistence to behave more defensively around write-time evidence storage.

v0.2.0

These changes shipped in the Phantasm v0.2.0 release.

Added

  • Added search.params.match_mode values tokens, exact, and fuzzy, with token matching as the default for natural multi-term queries.
  • Added search.params.rank_by with deterministic and relevance ordering modes.
  • Added search match diagnostics with per-result scores, matched fields, and matched terms.
  • Added search.params.group_by=record_kind aggregations with grouped counts and sample record IDs.
  • Added search.filters.payload_contains and search.filters.payload_matches for payload substring and regex filtering.
  • Added search.filters.subject_key, record_kind, created_after, and created_before for practical metadata filtering.
  • Added phm as a shorthand CLI binary for the same command surface as phantasm.
  • Added phantasm handle-request --stdin for request envelopes supplied through standard input.
  • Added phantasm handle-request --file <path> for UTF-8 request envelope files.
  • Added machine-readable cli and request_transport metadata to complete describe targets.
  • Added top-level response notices metadata for all successful handle-request responses.
  • Added first-call release_update notices after binary upgrades, scoped per bootstrapped project and client profile.
  • Added project-local release notice seen state using release_notice_seen.<client_profile>.
  • Added structured confirmation validation for high-impact operations. resolve_conflict, snapshot_import, backup_restore, and maintenance_run now require approved_by, reason, operation, and a target object matching the mutation request.
  • Added duplicate subject_key ingest refusal so callers get the existing record id and clear revise guidance instead of accidentally creating conflicting truth.
  • Added public and internal documentation for the phm alias, request input modes, response notices, release update notices, structured confirmations, ingest create-vs-update semantics, and release metadata maintenance.

Changed

  • Updated describe schemas, built-in help, public docs, and site docs to document search match modes, ranking, aggregations, and expanded filters.
  • Updated describe target=all|*|api to document response metadata, notices, release update notice shape, CLI aliasing, and request transport options.
  • Updated health guidance to make it a good first call after agent startup because it can surface one-time release notices.
  • Updated built-in help handle-request with the --stdin, --file <path>, and notices behavior.
  • Updated public install and usage docs to verify both phantasm and phm.
  • Updated ingest schema, built-in help, public docs, site docs, and managed agent guidance to make it explicit that ingest creates new memory and revise updates existing memory.
  • Updated conflict resolution examples to include the required structured confirmation payload.
  • Updated release documentation to require reviewing embedded release notice metadata before tagging.

Fixed

  • Fixed JSON parsing for printable UTF-8 strings so non-ASCII text is preserved correctly.
  • Fixed JSON \uXXXX escape handling, including valid surrogate pairs and invalid surrogate rejection.
  • Fixed JSON string output so printable Unicode remains readable while control characters are still escaped.

Earlier releases

Earlier public releases established the initial local project memory runtime, bootstrap flow, agent guidance management, JSON handle-request API, operation catalog, SQLite-backed project state, snapshots, backups, maintenance operations, and public install paths.