# Changelog ## [1.1.2] ### CodeGraph architecture documentation - Adds a shared CodeGraph Map illustration to the README and GitHub Pages, covering nine source module groups and 35 TypeScript files. - Saves a sanitized Map snapshot with declared relationships and a separately verified dynamic history import. - Separates source dependencies from historical-review participation and routing. - Documents module responsibilities, type-only dependencies, foundation boundaries, and explicit map refresh steps. - Adds source-to-snapshot and snapshot-to-SVG consistency checks, plus website map accessibility and navigation checks. - Packages the architecture SVG and evidence snapshot with the existing source architecture guide. - Preserves public tools, commands, configuration, and runtime behavior. Publishing and website deployment remain separate actions. ## [1.1.1] ### Domain organization - Groups source modules by history, settings, relevance, recall, knowledge, presentation, consultation, and setup responsibilities. - Extracts shared transcript and consultation record types into `src/contracts/` to remove reverse implementation dependencies. - Keeps `extensions/index.ts` as the Pi entry point and preserves existing tools, commands, and review behavior. - Updates internal imports, tests, and the packaged setup-guide path for the new directories. - Adds an architecture test for domain boundaries, import targets, and dependency cycles. - Documents module responsibilities and invocation relationships in the README and source architecture guide. - Excludes `.github/`, `scripts/`, `site/`, and `sites/` from the repository's CodeGraph index while retaining tests. - Leaves Git tracking, npm packaging, and SQLite transcript indexing unchanged. ## [1.1.0] ### Pi-only history - Removes Claude Code and Codex JSONL parsers and SQLite imports. - Removes Claude Code and Codex project-session migration and its Agent tool and command-menu action. - Aligns maintained history with Discussion, which currently activates participants from native Pi sessions. - Avoids maintaining external session formats that Discussion cannot activate directly. - Restricts session, source-file, retrieval, and project metadata types to native Pi history. - Deletes existing non-Pi rows when an older SQLite index opens. - Updates tests, setup guidance, package metadata, and website content for Pi-only support. ## [1.0.1] ### Package presentation - Highlights SQLite-backed search across scattered Pi, Claude Code, and Codex history, optional multi-agent analysis, and reusable evidence-based knowledge. - Adds npm keywords for session memory, cross-session history, multi-agent discussion, SQLite, and knowledge management. - Aligns the README introduction, website headline, and search and social descriptions with the package message. - Clarifies that retrieval uses literal substring matching, not typo-tolerant fuzzy search or vector similarity. - Emphasizes retrieving source evidence instead of relying on lossy summaries, while clarifying that the plugin complements rather than replaces Pi's context compaction. - Keeps retrieval, Discussion, and storage behavior unchanged. ## [1.0.0] Version 1.0 establishes the current product baseline. The previously published npm `latest` version was `0.7.23`. ### Project website - Adds an original dark memory-ledger design with amber highlights, rounded feature cards, and a review timeline. - Adds a responsive GitHub Pages landing page with feature summaries and setup instructions. - Uses separate static HTML, CSS, JavaScript, and SVG assets without a frontend framework. - Adds GitHub Actions deployment that uploads only the website directory. - Adds website and clipboard tests while keeping the unpublished package version at 1.0.0. - Presents independent Discussion and source-verification routing as a primary 1.0 architecture feature. - Adds a prominent responsive routing diagram to the hero and architecture section. - Highlights the multi-agent topology: one main agent and up to three user-confirmed historical Pi agents. - Shows direct transcript review, sequential private Discussion, independent `fetch_session` verification, and cited output as distinct paths. - Links the homepage to ADR 0001 for the complete decision and implementation evidence. ### Historical review workflow - Adds saved-knowledge-first review for explicit requests to look back at earlier work. - Adds description-only metadata inspection and exact saved-document reuse. - Runs global and entity-discovered project retrieval through one `recall_memory` call. - Requires a verbatim primary phrase for deterministic ranking and keeps global and project results separate. - Adds exact `turn_id` context expansion with explicit `before` and `after` windows. - Keeps System One filtering optional and provides a separate audit of filtered-out evidence. ### Discussion - Uses English labels in the historical-session picker and consultation results. - Keeps historical questions and replies in their original language. - Adds optional private multi-agent consultation with user-confirmed historical Pi sessions. - Runs historical participants sequentially in memory without modifying their source JSONL files. - Adds private follow-up, explicit finish, and explicit cancellation tools. - Keeps main-agent tool permissions unchanged and never treats participant agreement as verification. - Keeps `fetch_session` available for independent source checks during Discussion without automatic fallback or route replacement. - Documents the request-scoped routing contract, alternatives, and consequences in ADR 0001. ### OKF knowledge - Adds draft OKF v0.2 output for transcript and consultation evidence routes. - Stores portable meeting minutes, evidence files, an index, and an update log. - Reduces minutes and evidence frontmatter to one meaningful `description` property. - Requires findings, decisions, uncertainty, disagreements, open questions, and actions. - Returns the exact saved document as the final historical-review response. - Supports either the default knowledge root or one configured Obsidian vault as the storage root. ### Setup and configuration - Adds `setup_memory` for a sequential one-question-at-a-time setup guide. - Adds `configure_memory` for approved, validated, atomic configuration writes. - Preserves unrelated top-level settings and rejects malformed configuration without fallback. - Supports optional Discussion, System One filtering, managed or unmanaged endpoints, Obsidian storage, and a custom non-Obsidian knowledge root. - Keeps credentials in environment variables and excludes setup from history operations and service checks. ### Index safety and context isolation - Excludes current and retired package-tool turns from the searchable SQLite index. - Adds explicit cleanup with a verified SQLite backup and exact whole-turn deletion. - Preserves source JSONL, saved OKF files, session metadata, and source tracking during cleanup. - Keeps large tool evidence in temporary model context while persisting lightweight tool records. - Improves Pi branch ownership, incomplete-turn indexing, stable turn identities, and parser-version reindexing. ### Operations - Adds native current-project Claude Code and Codex migration with source selection. - Adds explicit history statistics and full backfill tools. - Limits `/pi-session-memory` to setup, history import, project-session migration, and indexed-tool-turn cleanup. - Removes Help, search, status, knowledge, consultation, and conclusion actions from the command menu. - Uses Pi's native extension update display instead of custom release-note state. ### Difference from npm 0.7.23 - Expands the registered Agent tools from 6 to 15. - Adds nine tools for setup, configuration saving, saved knowledge, Discussion, OKF finalization, and index cleanup. - Replaces transcript-only recall with a complete knowledge-first evidence and conclusion workflow. - Replaces the broad menu with four direct lifecycle operations; transcript review remains discoverable through Agent tools. - Replaces accumulated README version narratives with current product documentation and this single 1.0 baseline.