v0.3.30

Fixed

• Fixed dashboard operation detail logs so they honor the selected project instead of falling back to shared logs.
• Fixed project-scoped operations activity by paging through the session catalog until matching project runs are found.
• Hardened source-session catalog migrations and FTS rebuilds so older or malformed catalogs recover into a healthy indexed state for graph and source views.

v0.3.29

Fixed

• Fixed dashboard graph project switching by enforcing selected-project scope on graph responses and clearing stale graph data during project reloads.
• Made dashboard record, graph, artifact, operations, and source views fail closed instead of preserving stale rows after scoped load failures.
• Fixed the Records tab All statuses filter so it returns active and archived project records instead of falling back to the active-only default.
• Hardened the dashboard launcher against mismatched backend versions so an old local API cannot silently serve stale project data to a newer dashboard.
• Made invalid project selections return clear errors for status, memory, and Clinic endpoints, and kept refine reports degraded instead of crashing when the session catalog is unavailable.

Published artifacts:

• PyPI: lerim 0.3.29
• Docker: ghcr.io/lerim-dev/lerim:0.3.29

v0.3.28

Fixed

• Updated the offline release integration expectation for project-list count fields so the release workflow validates the new dashboard project counts.

Includes

• Dashboard project selection fixes and active/archived/total record count clarity from v0.3.27.

v0.3.27

Changed

• Made dashboard project selection visibly affect graph, records, brief, Clinic, source, operations, overview, and insights views, while keeping shared runtime surfaces clearly labeled.
• Separated current active, archived, and total persisted record counts across project lists, status payloads, and Run Clinic.

Fixed

• Fixed plain missing project names being interpreted as relative paths, so invalid dashboard project selections now return a clear error instead of falling back to the current workspace.
• Scoped graph queries, record filters, run detail reads, and artifact history through the selected project, including child workspace paths.
• Kept lerim serve and dashboard session-backed endpoints available in a degraded state when the source-session catalog is unhealthy.

v0.3.26

Added

• Added project selection to the dashboard surfaces that read project-scoped data, including overview, analytics, context, graph, traces, memory artifacts, operations, and Run Clinic.
• Added the Run Clinic dashboard artifact surface and project-scoped generated artifact history views.

Changed

• Kept selected projects in dashboard navigation URLs while leaving shared Skills and Settings views unscoped.
• Labeled shared runtime and multi-project operation rows clearly when they appear inside a project-scoped dashboard view.

Fixed

• Scoped dashboard status, queue, source-session, graph, artifact-history, and record-open reads to the selected project, including child workspace paths.
• Prevented cross-project artifact history and mixed-project operation totals from leaking into project-scoped dashboard views.

v0.3.25

Added

• Added README and docs guidance for registering skill targets, reviewing skill update proposals, and using opt-in auto-apply.

Changed

• Refined the Skills dashboard proposal review into a desktop queue/detail layout with explicit risk labels and cleaner applied-proposal state.
• Reduced normal dashboard tab gutters so pages use the available workspace with compact responsive padding.

v0.3.24

[0.3.24] - 2026-05-31

Added

• Added skill stewardship registration, refresh, proposal review, diff preview, and apply/reject flows across the CLI, API, and dashboard.
• Added scoped record loading, artifact-surface validation, and opt-in auto-apply policy controls for registered instruction targets.

Fixed

• Hardened skill proposal application so stale concurrent apply attempts cannot revert files after a proposal has already been marked applied.

v0.3.21

• Simplify Working Memory into a clearer short-term handoff.
• Reduce the default Working Memory recency window from 6 hours to 2 hours.
• Align Working Memory sections with Context Brief style: Summary, Start Here, Recent Changes, Current Context, and Open Questions.
• Refresh docs, tests, and version metadata for 0.3.21.

v0.3.20

• Refactor Working Memory generation into a first-class DSPy pipeline.
• Add structured Working Memory signatures and schemas with record-citation validation.
• Keep Working Memory status, freshness, paths, and manifest handling deterministic.
• Bump package version to 0.3.20.

v0.3.19

• Refactor agent runtime from BAML/LangGraph to DSPy 3.3.0b1 pipelines.
• Use MiniMax M2.7 defaults through the shared model runtime.
• Remove legacy CLI aliases and BAML/LangGraph code paths.
• Refresh docs, tests, and benchmark artifacts for the DSPy runtime.