Skip to content

CHANGELOG.md

Latest commit

 

History

History
621 lines (497 loc) · 31.1 KB

CHANGELOG.md

File metadata and controls

621 lines (497 loc) · 31.1 KB

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Added

  • [Semantic Search]: QMD integration for vector and hybrid search
    • New skill /mnemonic:qmd-setup for automated setup
    • New skill /mnemonic:qmd-reindex for re-indexing after captures
    • Supports BM25 keyword, vector semantic, and hybrid search modes
    • Auto-discovers memory roots from config
    • Registered collections: org-level, default, and project-level memories
    • See docs/semantic-search.md for details

Planned

  • Export/import functionality
  • Web UI for memory browsing

[0.6.1] - 2026-02-14

Changed

  • [MIF Spec Migration]: Replaced MIF git submodule with canonical mif-spec.dev schemas
    • Removed mif git submodule and .gitmodules
    • Bundled fallback ontologies are now the primary source (no submodule required)
    • All schema $id and schema_url references updated to https://mif-spec.dev/schema/
    • OntologyLoader resolution order simplified: bundled fallback -> user -> project
    • OntologyLoader exposes MIF_SCHEMA_URL and ONTOLOGY_SCHEMA_URL class constants

Removed

[0.6.0] - 2026-02-11

Added

  • [Capture Validation]: Ontology validation step in capture workflow

    • Namespace validation against loaded ontology registry
    • Memory type validation against ontology-defined types
    • Mandatory dedup check before memory creation

  • [Relationship Validation]: Type-safe relationship management

    • is_valid_type() guard in add_relationship() rejects unknown types
    • Bidirectional back-ref check rejects forward type for asymmetric relationships
    • is_symmetric() check ensures only symmetric types accept forward type as back-ref
    • Unknown relationship types in ensure_bidirectional() emit warnings instead of silently defaulting

  • [Relationship Type Registry]: Full MIF relationship type system (lib/relationships.py)

    • 9 core types with proper inverse mapping: RelatesTo, DerivedFrom, Supersedes, ConflictsWith, PartOf, Implements, Uses, Created, MentionedIn
    • RELATIONSHIP_TYPES registry with inverse, symmetric, and description metadata
    • get_inverse(), is_valid_type(), is_symmetric(), get_all_valid_types() helpers
    • to_pascal() / to_snake() case conversion between PascalCase and snake_case
    • Backward-compatible RECIPROCAL_TYPES dict with proper inverse types

  • [Team Audit Command]: /mnemonic:team-audit for cross-team memory health checks

  • [Test Coverage]: 397 tests (up from 308)

    • tests/unit/test_relationship_registry.py — Type registry, inverse mapping, case conversion (50+ tests)
    • tests/unit/test_ontology_validation.py — Namespace and type validation
    • tests/unit/test_memory_reader_yaml.py — YAML frontmatter parsing edge cases

Fixed

  • [Blank Line Accumulation]: Fixed add_relationship() reassembly that added extra blank lines on each write
  • [Namespace Validation]: Unknown sub-namespaces like _semantic/does_not_exist now correctly rejected even when top-level prefix matches
  • [Import Cleanup]: Removed unused imports and variables across lib and tests; resolved all ruff lint errors
  • [MIF Spec]: Updated to latest with ${MNEMONIC_ROOT} path references

Changed

  • [Hook Refactoring]: All hooks import from shared lib/ modules (continued from v0.5.0)
  • [Version Lineage]: Corrected version numbering to v0.x series

[0.5.0] - 2026-02-09

Added

  • [Shared Library Modules]: Extracted duplicated logic into reusable lib/ modules

    • lib/ontology.py — Unified ontology loading (get_ontology_file(), load_ontology_data(), load_file_patterns(), load_content_patterns(), load_ontology_namespaces(), get_fallback_file_patterns(), get_fallback_content_patterns(), get_ontology_info())
    • lib/search.py — Memory search and relationship inference (search_memories(), find_related_memories(), find_related_memories_scored(), find_memories_for_context(), detect_file_context(), detect_namespace_for_file(), extract_keywords_from_path(), extract_topic(), infer_relationship_type())
    • lib/memory_reader.py — Memory metadata extraction (get_memory_summary(), get_memory_metadata())
    • lib/relationships.py — Relationship write path and bidirectional linking (add_relationship(), add_bidirectional_relationship(), RECIPROCAL_TYPES)

  • [Relationship Write Path]: Programmatic relationship management

    • add_relationship() appends typed relationships to MIF frontmatter
    • add_bidirectional_relationship() creates forward + reciprocal back-links
    • Handles all frontmatter variants: no relationships section, block form, empty list
    • Duplicate detection prevents redundant entries

  • [Scoring & Relationship Constants]: Named constants replace magic numbers and strings

    • Relationship types: REL_RELATES_TO, REL_SUPERSEDES, REL_DERIVED_FROM
    • Scoring weights: SCORE_NAMESPACE_TYPE=30, SCORE_NAMESPACE_EXACT=20, SCORE_TAG_OVERLAP=20, SCORE_TITLE_KEYWORD=15, SCORE_CONTENT_KEYWORD=5, SCORE_MIN_THRESHOLD=15
    • All constants exported from lib/__init__.py

  • [Session-Scoped Blackboard]: Per-session blackboard isolation

    • PathResolver.get_session_blackboard_dir() for session-specific directories
    • PathResolver.get_handoff_dir() for cross-session handoff
    • migrate_blackboard_to_session_scoped() one-time migration from flat to session-scoped structure
    • Session metadata via _meta.json for programmatic discovery/cleanup
    • ADR-011 documents the architectural decision

  • [CLI Path Tool]: tools/mnemonic-paths for shell-based path resolution

    • Subcommands: root, blackboard, session-bb, search-paths, org, project
    • Replaces 6-line bash config resolution duplicated across markdown files

  • [Configurable Memory Store]: Memory store path is now user-configurable

    • Config stored at ~/.config/mnemonic/config.json (XDG-compliant, fixed location)
    • New lib/config.py module with MnemonicConfig class and canonical get_memory_root()
    • lib/paths.py PathContext gains memory_root field, resolved from config
    • hooks/session_start.py injects MNEMONIC_ROOT into session context
    • hooks/user_prompt_submit.py uses configured path for recall triggers
    • commands/setup.md prompts user for path, creates config, supports auto-migration
    • skills/mnemonic-setup/SKILL.md updated with config step and migration support
    • Default remains ~/.claude/mnemonic for backward compatibility

  • [Test Coverage]: 308 unit tests across all library modules

    • tests/unit/test_lib_ontology.py — Ontology loading and fallback patterns
    • tests/unit/test_search.py — Search, scoring, keyword extraction, relationship inference
    • tests/unit/test_memory_reader.py — Memory metadata and summary extraction
    • tests/unit/test_relationships.py — Hook integration for relationship suggestions
    • tests/unit/test_relationships_writer.py — Relationship write path (20 tests)

Changed

  • [Hook Refactoring]: All hooks now import from shared lib/ modules

    • hooks/pre_tool_use.py — Imports from lib.ontology and lib.search (228 → ~80 lines)
    • hooks/post_tool_use.py — Imports from lib.ontology and lib.search (252 → ~80 lines)
    • hooks/user_prompt_submit.py — Imports from lib.ontology, lib.search, lib.memory_reader (315 → ~100 lines)
    • hooks/session_start.py — Imports from lib.ontology (516 → ~200 lines)
    • hooks/stop.py — Session-scoped blackboard writes

  • [DRY Refactoring]: Eliminated duplicated code across hooks and search

    • STOPWORDS extracted as module-level frozenset (3 copies → 1)
    • _extract_keywords() private helper consolidates 4 instances of regex+filter pattern
    • Removed redundant get_memory_metadata() I/O in post_tool_use.py (metadata built inline from scored search results)
    • ~165 lines of duplicated path resolution code removed

  • [Unified Path Alignment]: All components now use $MNEMONIC_ROOT with config resolution

    • Replaced all hardcoded ~/.claude/mnemonic and $HOME/.claude/mnemonic across 30 files
    • Removed all ./.claude/mnemonic project-local path references (unified structure has no project-local dir)
    • Added config resolution blocks to all bash-based commands, skills, and agents
    • commands/status.md rewritten to show store/org/project directories under unified path
    • tools/mnemonic-query delegates to PathResolver instead of reimplementing paths
    • tools/mnemonic-validate uses get_memory_root() for search path resolution

  • [Project Detection]: Improved 3-tier project name detection

    • lib/paths.py _detect_project() now tries: git remote → git toplevel → cwd name
    • Same pattern applied to hooks/session_start.py, commands/capture.md, commands/recall.md
    • Prevents wrong project name when running from subdirectories

  • [Claude Code Identity]: Plugin is now exclusively a Claude Code plugin

    • README rewritten to remove 9 multi-tool badges and integration table
    • Enterprise docs updated to reference Claude Code only
    • Multi-tool integration guides archived to docs/archive/
    • Community docs (Memory Bank migration/comparison) archived to docs/archive/
    • Templates (Cursor, Windsurf, Copilot, Codex) archived to docs/archive/

Removed

  • [Multi-Tool References]: Removed all non-Claude-Code tool references
    • .github/copilot-instructions.md — Deleted (Copilot-specific configuration)
    • docs/integrations/ — Archived (Cursor, Windsurf, Aider, Continue, Codex CLI, Gemini CLI, OpenCode, GitHub Copilot guides)
    • docs/community/ — Archived (Memory Bank migration, comparison, quickstart, adoption stories)
    • templates/ — Archived (AGENTS.md, CONVENTIONS.md, copilot-instructions.md, cursor-rule.mdc, mnemonic-protocol.md, plugin-hooks/)

[0.4.0] - 2026-01-30

Added

  • [Custodian Skill]: Unified memory maintenance command (/mnemonic:custodian)

    • audit - Full health check: frontmatter, links, decay, relationships, orphans
    • validate-links --fix - Check and repair broken wiki-links and UUID references
    • validate-memories - MIF schema compliance for frontmatter fields
    • validate-relationships - Ontological relationship type and target validation
    • decay - Recalculate exponential decay strength values in-place
    • summarize - Find memories eligible for compression-worker agent
    • relocate <old> <new> - Move memories with cross-reference updates and git mv
    • Options: --dry-run, --fix, --json, --commit
    • 8 Python modules in skills/custodian/lib/

  • [Filename Migration]: Auto-migration utility in lib/migrate_filenames.py

    • Renames UUID-prefixed files ({uuid}-{slug}.memory.md) to slug-only ({slug}.memory.md)
    • Merges content when slug collisions occur (atomic writes via temp files)
    • Idempotent: migration marker prevents re-scanning on subsequent sessions
    • CLI support: --dry-run for preview, --force to re-run

  • [Ontology Discovery]: Enhanced pattern-based entity discovery (f642fde)

    • Improved schema validation for custom ontologies
    • Discovery patterns for automatic entity detection

  • [Integration Skill]: Python-based memory integration with test suite (1681803)

    • Migrate memories between organizations/projects
    • Batch operations with progress tracking
    • Comprehensive test coverage

  • [Memory Protocol]: Stronger prompt engineering for compliance (ba39754)

    • Commitment-based framing (vs command-based)
    • Identity reinforcement for persistent memory
    • Stop hook blocks until captures complete

  • [Social Graphics]: PNG versions for README badges (cfa77af)

    • Cognitive namespace visualizations
    • Ontology structure diagrams

  • [MIF Spec]: Ontologies aligned with MIF

    • mif-base.ontology.yaml defines cognitive triad hierarchy
    • JSON-LD export for semantic web compatibility
    • scripts/yaml2jsonld.py converter

  • [Cognitive Triad Namespaces]: Migrated from 9 flat namespaces to hierarchical structure

    • Old: apis, blockers, context, decisions, learnings, patterns, security, testing, episodic
    • New: _semantic/, _episodic/, _procedural/ with sub-namespaces
    • _semantic/decisions - Architectural choices
    • _semantic/knowledge - APIs, context, learnings, security
    • _semantic/entities - Technologies, components
    • _episodic/incidents - Production issues
    • _episodic/sessions - Debug sessions
    • _episodic/blockers - Impediments
    • _procedural/runbooks - Operational procedures
    • _procedural/patterns - Code conventions, testing
    • _procedural/migrations - Migration steps

  • [Ontology Loader]: Centralized MIF loading (skills/ontology/lib/ontology_loader.py)

  • [Fallback Support]: Offline installations via skills/ontology/fallback/

  • [Namespace Migration]: scripts/migrate_namespaces.py for upgrading existing memories

  • [Path Resolution Library]: Centralized path management in lib/paths.py

    • PathResolver class with unified path scheme
    • PathContext dataclass for context detection (org, project, home, scheme)
    • PathScheme enum (LEGACY, UNIFIED) for migration support
    • Scope enum (USER, PROJECT, ORG) for memory scoping
    • Convenience functions: get_memory_dir(), get_search_paths(), get_blackboard_dir()
    • 26 unit tests in tests/unit/test_paths.py

  • [Migration Script]: scripts/migrate_to_v2_paths.py

    • Migrates memories from dual-location (LEGACY) to unified (V2) path structure
    • Automatic backup creation before migration
    • Rollback capability
    • --dry-run mode for previewing changes
    • Progress tracking and git commit of changes

Changed

  • [File Naming]: Memory files now use slug-only format ({slug}.memory.md)

    • UUID removed from filename; stored only in frontmatter id: field
    • Wiki-links [[slug-name]] now resolve correctly in Obsidian and similar tools
    • Existing UUID-prefixed files auto-migrated on session start
    • Capture command merges content when target file already exists
    • Updated across 21 files: commands, hooks, tests, templates, and documentation

  • [Unified Path Structure]: All memories now stored under ~/.claude/mnemonic/

    • Eliminates "split brain" between user-level and project-level storage
    • New structure: {org}/{project}/{namespace}/ for project-specific memories
    • Org-wide: {org}/{namespace}/ for memories shared across projects
    • Fallback: default/{namespace}/ when org detection fails
    • Single git repository for all memories

  • [Cognitive Namespace Prefix]: All cognitive triad namespaces now prefixed with _ (40752d1)

    • _semantic/, _episodic/, _procedural/ distinguish system namespaces
    • Prevents collision with user-defined namespaces

  • [Documentation]: Migrated to cognitive namespace structure (bcfe79f)

    • All docs updated for unified path structure
    • Social graphics reflect new ontology model

  • [Hooks]: Updated to use PathResolver and hierarchical namespace paths

    • session_start.py: Uses PathResolver for path resolution with LEGACY fallback
    • user_prompt_submit.py: Uses PathResolver for memory search paths
    • Hooks check both unified and LEGACY paths during migration period

  • [Commands]: Updated scope options and namespace format

    • capture.md: --scope project|org (was user|project), supports new namespace format
    • recall.md: --scope project|org|all (was user|project|all)

  • [Code Quality]: Simplified functional test infrastructure

    • Extracted ClaudeRunner and MemoryHelper classes to module level
    • Extracted NAMESPACE_MAP and MEMORY_WRITE_DELAY constants
    • Replaced debug prints with logging.debug()
    • Simplified verbose GIVEN/WHEN/THEN docstrings to concise single-line format
    • Improved assertion messages with expected vs actual context

  • [Scripts]: Improved migration and repair scripts

    • migrate_scope_paths.py: Returns (migrations, errors) tuple for better error handling
    • fix_malformed_memories.py: Extracted _format_yaml_field helper, replaced conditional chains with dict mapping
    • Removed unused current_key variable from YAML parser

  • [Hooks]: Refined user_prompt_submit.py

    • Simplified detect_triggers() using any() pattern
    • Use specific exception types (yaml.YAMLError, OSError, subprocess.SubprocessError)

Fixed

  • [Hooks]: Read tool data from stdin instead of environment variables (39cf67a)

    • Fixes hook data passing in Claude Code environment
    • All hooks now properly receive tool context

  • [Plugin]: Removed duplicate hooks reference from manifest (b2deea3)

    • Hooks now loaded exclusively from hooks/hooks.json
    • Added .fastembed_cache/ to gitignore

  • [Security]: Hardened ontology and integrate skills (a3dcc4a)

    • Improved input validation
    • Better error handling for edge cases

  • [Git History]: Purged .fastembed_cache/ directories from repository history

Migration Guide

To migrate namespaces (one-time, required):

python scripts/migrate_namespaces.py --dry-run
python scripts/migrate_namespaces.py --commit

To migrate to unified path structure:

# Preview changes
python scripts/migrate_to_v2_paths.py --dry-run

# Execute migration (creates backup automatically)
python scripts/migrate_to_v2_paths.py

# Rollback if needed
python scripts/migrate_to_v2_paths.py --rollback ~/.claude/mnemonic_backups/legacy_backup_TIMESTAMP

[0.3.0] - 2026-01-25

Changed

  • [CLAUDE.md Protocol]: Role-based, imperative instructions (Anthropic Prompt Engineering)

    • Added role assignment: "developer with persistent memory"
    • Changed passive "Ask yourself" to structured <capture_eval> block
    • Reduced from 63 lines to 30 lines for focused instruction
    • Imperative language throughout ("Search first. Always.")

  • [mnemonic-core Skill]: Progressive disclosure architecture

    • Reduced SKILL.md from 495 lines to 52 lines (quick reference only)
    • Split detailed workflows into references/ subdirectory:
      • capture.md - Full capture workflow
      • recall.md - Search and retrieval patterns
      • schema.md - Complete MIF Level 3 schema
      • examples.md - Working examples
    • Follows Anthropic Architect principle: "Show only what's needed, when it's needed"

  • [mnemonic-format Skill]: Simplified schema emphasis

    • Lead with minimal 4-field template (id, title, type, created)
    • All other fields marked as optional
    • Reduces barrier to adoption

  • [Hook Architecture]: Context hints only, no auto-capture

    • user_prompt_submit.py: Removed create_memory() function
    • Hooks provide capture signals, Claude uses skills to capture
    • Aligns with skills-first architecture principle
    • post_tool_use.py: Updated to reference /mnemonic:capture skill

  • [Commands]: Updated CLAUDE.md templates in init.md and setup.md

    • Match new role-based, imperative style
    • Simplified initial context memory to 4-field format

  • [Test Suite]: Replaced contrived tests with natural behavior tests

    • Removed 25+ "no error" command tests
    • Added 18 functional tests validating automatic capture/recall without explicit commands
    • Tests verify silent operation, content quality, no duplicates
    • Tests use --plugin-dir flag to properly load plugin hooks

Added

  • [Stop Hook Enforcement]: Blocks Claude from stopping until captures complete

    • stop.py checks for pending captures via temp file
    • Uses decision: "block" with reason to force capture before stopping
    • Prevents stop_hook_active infinite loops
    • Clears pending state after successful capture

  • [Pending Capture State]: Session-aware capture tracking

    • user_prompt_submit.py writes pending captures to /tmp/mnemonic-pending-{session_id}.json
    • Uses CLAUDE_SESSION_ID for consistent cross-hook communication
    • Stop hook reads and enforces pending captures

Fixed

  • [Pattern Triggers]: Fixed pattern namespace detection

    • Added \bshould always\b and \balways use\b patterns
    • Original \balways\b.*\bwhen\b missed common phrases like "We should always use X"

  • [Skill Description]: Updated mnemonic-core for auto-loading

    • Added explicit trigger phrases to description field
    • Enables Claude to auto-load skill when relevant phrases detected

  • [Templates]: Heredoc variable expansion in mnemonic-core/SKILL.md

    • Changed 'MEMORY_EOF' to MEMORY_EOF (unquoted)
    • Changed {UUID} placeholders to ${UUID} bash variables

Removed

  • [Cognitive Bridge]: Entire cognitive bridge feature removed
    • Removed bridge commands: /mnemonic:bridge-scan, /mnemonic:bridge-status, /mnemonic:bridge-sync
    • Removed bridge skills: bridge-core, bridge-triggers, bridge-synthesis, bridge-integration
    • Removed bridge agents: capability-mapper, pattern-synthesizer
    • Removed check_registry() from session_start.py hook
    • Removed bridge documentation and architecture diagrams
    • Skills-first approach replaces bridge's hook-centric pattern

[0.2.0] - 2026-01-24

Added

  • [Cognitive Bridge]: Plugin integration and intelligent context injection

    • PreToolUse hook for file pattern detection (auth, api, db, test, config, deploy, security)
    • Relevant memory file paths provided when editing related files
    • Bridge registry for plugin capability mapping
    • Pattern synthesis for cross-memory knowledge consolidation

  • [Bridge Skills]: Four new skills for cognitive bridge

    • bridge-core: Interpreting session context and memory availability
    • bridge-triggers: File pattern to namespace mappings
    • bridge-synthesis: Pattern consolidation and meta-pattern creation
    • bridge-integration: Plugin integration for memory capture

  • [Bridge Agents]: Two new agents for automated operations

    • capability-mapper: Discovers plugins and builds bridge registry
    • pattern-synthesizer: Analyzes memories for patterns and creates meta-patterns

  • [Bridge Commands]: Three new commands

    • /mnemonic:bridge-scan: Discover plugins and build bridge registry
    • /mnemonic:bridge-status: Show registry, memory health, blackboard status
    • /mnemonic:bridge-sync: Process blackboard and synthesize patterns

Changed

  • [Hooks]: All hooks now use hookSpecificOutput.additionalContext for Claude-readable context

    • SessionStart: Provides memory counts, health score, registry status, suggestions
    • PreToolUse: Provides relevant memory file paths for file edits
    • UserPromptSubmit: Provides capture/recall trigger context
    • PostToolUse: Provides capture opportunity context
    • Stop: Provides session summary context

  • [Architecture]: Hooks inform, Claude decides

    • Hooks provide context, not instructions
    • Claude autonomously decides when to read memories or use agents
    • Agents invoked via commands, not hook triggers

Technical

  • Hooks execute in <100ms with timeouts (3-5s)
  • Fast pattern matching with regex and ripgrep
  • Memory health scoring based on confidence decay and duplicates

[0.1.1] - 2026-01-24

Added

  • [Query Command]: New /mnemonic:query command for confidence-rated memory retrieval
    • tools/mnemonic-query Python CLI with structured output
    • Confidence scoring based on age, strength, and access patterns
    • JSON output format for programmatic use
    • Test fixtures for confidence scoring validation
  • [Test Framework]: Extended test definitions in .claude/tests/tests.yaml
    • Additional test cases for query command
    • Improved test coverage
  • [Research Reports]: Comprehensive market research report
    • Full report in Markdown and HTML formats
    • Executive summary
    • Mermaid visualizations for competitive positioning, SWOT, risk matrix

Changed

  • [Search Skill]: Enhanced mnemonic-search with confidence-aware patterns
  • [README]: Updated feature documentation
  • [Makefile]: Added query-related targets

Fixed

  • [Reports]: Corrected inaccurate feature status in research report
    • Marked decay/archival as implemented (gc command)
    • Marked citation validation as implemented (mnemonic-validate)
    • Fixed Mermaid SWOT quadrant syntax (<--> → -->)
  • [Plugin]: Removed duplicate hooks reference from manifest

[0.1.0] - 2026-01-24

Added

  • [Citations]: Optional citations array in MIF format for external references
    • Supports types: paper, documentation, blog, github, stackoverflow, article
    • Fields: title, url, author, date, accessed, relevance, note
    • Validation rules for citation fields
    • Search patterns for finding memories by citation
  • [Enhanced Search]: Agent-driven iterative search with synthesis
    • /mnemonic:search-enhanced command for comprehensive memory analysis
    • mnemonic-search-enhanced skill for orchestrating multi-round search
    • mnemonic-search-subcall agent (Haiku) for targeted search operations
    • Iterative query refinement (up to 3 rounds)
    • Synthesized answers instead of raw file matches
  • [GitHub Integration]: GitHub Copilot agent configuration
  • [Plugin Config]: ADR and sigint plugin configurations
  • [Validation Tool]: MIF Level 3 schema validation for memory files
    • tools/mnemonic-validate Python CLI for validating memories
    • Validates required fields, types, UUID format, ISO 8601 timestamps
    • Validates code_refs (file paths, line numbers, types)
    • Validates citations (types, URLs, relevance scores)
    • Validates memory links ([[uuid]] resolution)
    • JSON and Markdown output formats
    • --capture flag to save validation runs as episodic memories
    • --changed flag for git-modified files only
    • Makefile targets: validate-memories, validate-memories-ci
  • [Memory Compression]: Auto-summarization for verbose memories
    • compression-worker agent (Haiku) for generating concise summaries
    • New gc flags: --compress, --compress-only, --compress-threshold
    • Compression criteria: (Age > 30d AND lines > 100) OR (strength < 0.3 AND lines > 100)
    • Adds summary and compressed_at fields to MIF frontmatter
  • [Multi-Agent Coordination]: Blackboard-based agent coordination
    • mnemonic-agent-coordination skill with agent patterns
    • Agent registration in session-notes.md
    • Task handoffs with context in active-tasks.md
    • Shared workflow state in shared-context.md
    • Progress reporting for subcall agents
    • Extended blackboard entry format with Agent, Status, Capabilities fields
  • [ADRs]: Five new architectural decision records
    • ADR-003: Agent Coordination via Blackboard Extension
    • ADR-004: MIF Schema as Single Source of Truth for Validation
    • ADR-005: Memory Compression via GC Extension
    • ADR-006: Validation Results as Episodic Memories
    • ADR-007: Enhanced Search with Agent-Driven Iteration
  • [Test Framework]: Hook-driven test framework for plugin validation
    • .claude/tests/tests.yaml - Test definitions with expectations
    • .claude/tests/runner.sh - Test execution and validation
    • .claude/hooks/test-wrapper.sh - User prompt interception for test mode
    • 25 functional tests covering commands, skills, agents, integration, and error handling

Changed

  • [Hooks]: Format Python files with ruff
  • [ADRs]: Rename ADRs to 3-digit numbering scheme
  • [GC Command]: Extended with compression capabilities
  • [Agents]: memory-curator and search-subcall now use blackboard coordination
  • [Skills]: mnemonic-search-enhanced uses blackboard for workflow state

Fixed

  • [Stop Hook]: Skip verbose output during test mode to prevent test interruption
    • Removed XML-style <mnemonic-session-end> tags that could cause parsing issues
    • Added is_test_mode() detection based on test-state.json
  • [Test Expectations]: Updated agent_search_subcall_basic test to use specific failure patterns

Documentation

  • [Reports]: Add AI memory filesystem research
  • [validation.md]: Validation tool usage and CI integration
  • [agent-coordination.md]: Multi-agent coordination patterns and examples
  • [Enterprise Package]: Comprehensive documentation for enterprise adoption
    • docs/enterprise/README.md - Overview and audience-specific navigation
    • docs/enterprise/compliance-governance.md - Audit trails, data sovereignty, SOC2/ISO/GDPR
    • docs/enterprise/productivity-roi.md - Team benefits, cost analysis, multi-tool flexibility
    • docs/enterprise/developer-experience.md - Privacy-first design, power user features
    • docs/enterprise/research-validation.md - Letta benchmark (74%), academic foundations
    • docs/enterprise/deployment-guide.md - Installation, backup, team sharing patterns
  • [Community Package]: Resources for Memory Bank users and community
    • docs/community/README.md - Community hub and navigation
    • docs/community/quickstart-memory-bank.md - 5-minute setup for Memory Bank users
    • docs/community/migration-from-memory-bank.md - Complete migration walkthrough
    • docs/community/mnemonic-vs-memory-bank.md - Side-by-side feature comparison
    • docs/community/adoption-stories.md - Community experiences template
    • docs/community/CONTRIBUTING-COMMUNITY.md - How to contribute stories
  • [Integration Guides]: Enhanced Cursor, Windsurf, GitHub Copilot guides
    • Common workflows and daily development patterns
    • Advanced configuration patterns
    • Memory Bank migration sections
    • Troubleshooting guides

[0.0.1] - 2026-01-24

Added

  • Initial release of Mnemonic memory system
  • Commands:
    • /mnemonic:setup - Configure mnemonic with proactive behavior
    • /mnemonic:init - Initialize directory structure
    • /mnemonic:capture - Capture a new memory
    • /mnemonic:recall - Search and recall memories
    • /mnemonic:search - Full-text search
    • /mnemonic:status - Show system status
    • /mnemonic:gc - Garbage collect expired memories
  • Skills:
    • mnemonic-core - Complete memory operations
    • mnemonic-setup - CLAUDE.md configuration
    • mnemonic-search - Advanced search patterns
    • mnemonic-format - MIF Level 3 templates
    • mnemonic-organization - Namespace management
    • mnemonic-blackboard - Cross-session coordination
  • Agents:
    • memory-curator - Autonomous maintenance and curation
  • Hooks:
    • SessionStart - Load relevant memories on session start
    • UserPromptSubmit - Detect recall/capture opportunities
    • PostToolUse - Capture learnings from tool results
    • Stop - Commit changes and summarize session
  • MIF Level 3 compliant memory format
  • Bi-temporal tracking (valid time vs recorded time)
  • Exponential decay model for memory relevance
  • Git versioning for all memories
  • Cross-session coordination via blackboard
  • User-level and project-level memory scopes
  • Nine standard namespaces (apis, blockers, context, decisions, learnings, patterns, security, testing, episodic)

Technical Details

  • Pure filesystem-based storage (no external databases)
  • Self-contained skills (no library dependencies)
  • Standard Unix tools (git, ripgrep, find)
  • Python 3.8+ for hook scripts