v1.18.0.0

Added

• read-efficient plugin (v1.1.1.0): New plugin for machine-efficient file reading
  • Token-efficient file reading for Claude Code with 30-70% token reduction
  • Converts files to compact JSON and removes redundant whitespace
  • Supports 10+ formats: JSON, CSV, YAML, XML, HTML, Markdown, SQL, logs, and plain text
  • Format-specific parsing: Markdown anchor line extraction, SQL AST parsing, log file auto-detection
  • Property-level minification with intelligent type conversion ("true" → boolean, "123" → number)
  • Single-file and batch processing with optional disk caching
  • 533+ comprehensive tests validating all parsers and edge cases
  • Zero information loss with graceful fallback to plaintext on parse errors

Fixed

• smart-compact plugin (evolved to v2.2.3.0): Critical bug fix
  • Fix division error in block_idle_session.py hook with inverted logic and None safety check
  • Prevents "unsupported operand type(s) for /: 'NoneType' and 'int'" error

Changed

• smart-compact plugin (evolved from v2.2.1.0 to v2.2.3.0): Code organization and bug fixes

  • v2.2.2.0: Moved configuration and duplicate logic into const_models_and_config.py to eliminate duplication across hook files
  • v2.2.3.0: Fixed critical division error in idle session blocker and added None safety checks

• fetch-full-content plugin (evolved to v1.2.1.0): Improved output formatting

  • Warning output when hidden content is filtered remains in stdout but removed from markdown file to reduce duplicate tokens

• project-intel plugin (evolved to v1.6.0.0): Enhanced scan batching

  • Scan output batching: Files are now pre-batched in scanner (8 files per batch) instead of requiring model to create batches
  • Scan.json structure: filesToScan is now array of arrays (batches) instead of flat array
  • Reduces model overhead and simplifies slash command implementation

---

Full Changelog: v1.17.0.0...v1.18.0.0

v1.17.0.0

Added

• smart-compact plugin evolved from v1.4.1.0 to v2.2.1.0:
  • v2.0.0.0: Complete algorithm rewrite from backward-iterating to forward-chaining for improved clarity and efficiency
    • Forward-chaining dedup: Processes reads chronologically per file for intuitive duplicate detection
    • Keep-last strategy: Latest occurrence represents current state (aligns with LLM token reading preference)
    • Content type auto-detection: Line-level diffing for multiline, character-level for single-line (JSON)
    • Configurable thresholds: Min bytes, context margins via environment variables
    • Raw content extraction: Detects Write → Read edge cases
    • Improved output formats: Short format (Savings: XXX bytes (~YYY tokens)) and normal reporting
  • v2.1.0.0: Extended deduplication to bash and grep operations
    • Bash dedup: Treats file-reading commands (cat, head, tail, wc) like Read operations
    • Grep dedup: Safely deduplicates grep searches when later reads exist
    • Unified operation stream: Process all operations (Read, Write, Bash, Grep, Edit) chronologically
    • Edit tracking: Validates dedup safety between grep and reads
  • v2.1.1.0: Smart edit overlap detection and extended bash patterns
    • Smart overlap: Only skip dedup if edits touch exact lines matched by grep (no false negatives)
    • Extended patterns: Support head, tail, wc with flags; bash -c wrapped commands
    • Line number extraction: Parse grep output to enable safe overlap checking
  • v2.2.0.0: Bash script execution deduplication
    • Script output dedup: Detect identical python, npm, node, dotnet, ruby, java, go outputs
    • Keep-last for scripts: Earlier identical runs marked redundant, latest output kept
    • Pattern recognition: Auto-identify script invocations (python script.py, npm test, etc.)
    • Context-aware markers: Separate messages for file reads vs script outputs
  • v2.2.1.0: Code restructuring and maintainability improvements
    • Package refactoring: Split 1299-line monolithic script into focused Python package
    • Modular structure: const_models_and_config, content, detection, extract, init modules
    • 72% complexity reduction in main script (1299 → 368 lines)
    • Single responsibility per module: Easier to test, extend, understand
    • Backward compatibility: All tests pass (edge cases, smart overlap, pattern matching)

Changed

• smart-compact internal structure: Reorganized from monolithic script to modular package

  • const_models_and_config.py (141 lines): Data models, enums, configuration, markers
  • content.py (197 lines): Content type detection and diffing
  • detection.py (106 lines): Pattern detection for grep, bash, edit operations
  • extract.py (377 lines): Transcript I/O and operation extraction
  • __init__.py (71 lines): Clean public API with all exports
  • Enhanced maintainability without changing external behavior or functionality

• Reduced allowed tools scope: Restrict tools to minimum necessary for each skill/command

  • changelog plugin: Removed allowed-tools from managing-changelog skill (informational only, auto-allows tools)
  • documentation plugin: Removed allowed-tools from managing-documentation skill (informational only, auto-allows tools)
  • fetch-full-content plugin: Restricted Bash(python:*) → Bash(python ${CLAUDE_PLUGIN_ROOT}/scripts/fetch_full_content.py *) (specific script only)
  • project-intel plugin: Restricted Bash(node:*) → Bash(node ${CLAUDE_PLUGIN_ROOT}/scripts/dist/ctx.js *) (specific script only)
  • session-protocol plugin: Restricted Bash → Bash(git rev-parse *) (git-only operations)
  • Improved security and predictability by preventing unintended tool usage

---

Full Changelog: v1.16.0.0...v1.17.0.0

v1.16.0.0

Added

• smart-compact plugin (evolved from v1.0.0.0 to v1.4.1.0): Major improvements to deduplication and user notifications
  • User notification system: Stop hook displays duplicate summaries (e.g., "Duplication in conversation: 15.4K characters (14.5K tokens, 7.2% of total context window)")
  • Configurable notification and cache validation thresholds via environment variables
  • Cross-platform system notifications via cross-platform-notification plugin (Windows, macOS, Linux)
  • Self-documenting deduplication markers for better LLM understanding
  • Human-readable formatting for bytes (KB, MB) and tokens (K)
  • Token estimation calculations improved (now uses standard ~4 bytes ≈ 1 token conversion)
  • Byte counting for resumed sessions now uses actual message content for accuracy

Changed

• Plugin infrastructure and configuration
  • Updated all plugin.json files and cleaned up marketplace.json for consistency
  • Fixed tsConfig in project-intel project for proper build configuration

Fixed

• project-intel plugin (v1.5.2.0): Scan command refinement
  • Fixed scan output example format in /scan slash command (prevented example output from overriding subagent format)

---

Full Changelog: v1.15.0.0...v1.16.0.0

v1.15.0.0

Added

• transcript-duplicate-scrubber plugin (v1.0.0.0): New plugin for automatic transcript deduplication
  • Hash-based duplicate detection (SHA256) for file content comparison
  • Two-level deduplication strategy: Write priority (Rule 1) and token priority for Reads (Rule 2)
  • Integration modes: Automatic via SessionEnd hook or manual CLI with --dry-run preview
  • Support for minified JSONL and pretty-printed JSON transcript formats
  • Unicode path support for non-ASCII characters
  • Detailed reporting with transparency markers for removed duplicates
  • O(n) time complexity with no false positives, reversible changes

---

Full Changelog: v1.14.0.0...v1.15.0.0

v1.14.0.0

Changed

• project-intel plugin (v1.4.0.0): Stability improvements

  • Fixed entry ordering when loading and saving from summaries
  • Fixed merge logic to avoid obsolete entry deletion
  • Improved ignore patterns to include dotfiles with selective filtering for build/cache directories
  • Fixed summary lookup map building during merge
  • Fixed file modification date comparison for accurate staleness detection

• project-intel plugin (v1.5.0.0): Code organization and discovery logic improvements

  • Extracted constants and types into dedicated types.ts file for better maintainability
  • Fixed knowledge directory discovery with early exit when location/process directory is higher in structure
  • Fixed unused calledFromHook early exit logic
  • Fixed git detection to check for git repo in location instead of only checking if git is installed
  • Simplified file system search logic in findKnowledgeDir for better reliability

• project-intel plugin (v1.5.1.0): Bug fixes for partial summaries and record structure

  • Fixed error when knowledgeDir argument wasn't provided on merge
  • Fixed remaining usages of old files and directories record structure, now uses map structure consistently
  • Corrected total sizes returned on scan/merge operations
  • Fixed deletion of obsolete directory entries

• session-protocol plugin (v1.2.0.0): Command simplification and improved session workflow

  • Breaking: Removed /save-session-protocol and /load-session-protocol slash commands - managing-session-continuity skill can be invoked directly if necessary
  • Improved assistant instruction in session start hook: complete workflow provided when session protocol exists

---

Full Changelog: v1.13.0.0...v1.14.0.0

v1.13.0.0

Changed

• project-intel plugin (v1.2.0.0): Session automation and intelligent knowledge maintenance

  • SessionStart hook with automatic notifications (total files, files needing update, scan suggestions)
  • Automatic deletion cleanup: Files and directories removed from scan location are cleaned from summaries
  • Filesystem modification date detection for parity with git-based change detection
  • Dynamic knowledgeDir discovery with intelligent fallback chain (explicit arg → location-based → process-based)
  • Eliminates need to repeatedly specify --knowledgeDir for same-directory operations

• project-intel plugin (v1.3.0.0): Query optimization with improved output structure and per-file technologies

  • Per-file technologies tracking: Each file captures its technologies (TypeScript, Python, JSON, etc.) during scan for granular matching
  • Optimized query output: Removed redundant path field in grouped format, now uses fileName + folderPath to eliminate duplication
  • Removed lastUpdated from query results (now irrelevant with SessionStart hook showing staleness)
  • Changed default query format to grouped (hierarchical organization by directory) for better architectural context
  • Flat format includes per-file technologies for detailed cross-file comparisons
  • Improved documentation: Clear guidance on when to use grouped vs flat output with use-case descriptions

• fetch-full-content plugin (v1.2.0.0): Security enhancements with prompt injection filtering

  • Basic prompt injection filtering for common hidden content vectors (comments, visibility, font-size, opacity, color alpha)
  • Warning output when hidden content is filtered (both stdout and markdown file)
  • Enhanced security notice clarifying filtering scope and limitations

---

Full Changelog: v1.12.0.0...v1.13.0.0

v1.12.0.0

Changed

• project-intel plugin (v1.1.0.0): Performance enhancement with Git-based incremental scanning
  • Only re-analyze files modified since last scan (80-95% reduction on subsequent scans)
  • Automatic git tracking detection with filesystem fallback
  • Modification timestamp tracking from git history for accurate staleness detection

---

Full Changelog: v1.11.0.0...v1.12.0.0

v1.11.0.0

Added

• project-intel plugin (v1.0.0.0): Lightweight reconnaissance system that provides direction before exploration
  • /scan command: Generate semantic summaries of project files and directories (one-time setup)
  • /query command: Search summaries by semantic relevance before reading files
  • Semantic scoring system matching on purpose, summary, exports/imports, technologies, and path
  • Wave-based parallel processing: Max 10 concurrent agents for large projects
  • Persistent storage: .knowledge/summaries.json survives across sessions
  • Directory and file summary generation with purpose, role, exports, imports, and technologies tracking
  • File modification timestamp tracking for staleness detection
  • Scope filtering: Limit queries to specific directories with --scope parameter
  • Result limiting: Control query results with --max parameter
  • Incremental updates: Re-scan merges new summaries with existing knowledge
  • Best practices guide for maintenance strategy and query approaches
  • Performance characteristics: Cost estimates for projects of different sizes
  • Troubleshooting guide with common issues and solutions
  • Comparison to alternatives: vs. Explore agent, Grep/Glob, direct file reading
  • Design decision documentation: Why persistent storage, wave-based processing, semantic scoring, query-first workflow, and user-managed staleness
  • Support for semantic matching on conceptual terms, not just keywords
  • Project structure guidance: When to use project-intel, when to skip it, breakeven analysis

---

Full Changelog: v1.10.0.0...v1.11.0.0

v1.10.0.0

Added

• documentation plugin (v1.0.0.0): New plugin for creating and maintaining high-quality project documentation
  • managing-documentation skill with 4 core workflows (WF1-WF4) for documentation lifecycle management
  • WF1: Creating New Documentation - define purpose, plan structure, write for global audience, ensure inclusivity, validate quality
  • WF2: Updating Existing Documentation - assess state, prioritize improvements, update content, verify changes
  • WF3: Validating Documentation Quality - check content quality, clarity, inclusivity, completeness
  • WF4: Applying Agile/Lean Principles - document late, update constantly, with purpose, choose best medium
  • Support for multiple documentation types: guides, API docs, README files, architecture documentation
  • ARID principles: Accept Repetition, Skimmable, Exemplary, Consistent, Current
  • Global audience support: simple language, defined abbreviations, concrete examples, no idioms
  • Inclusive language standards: no ableist, gendered, violent, or culturally specific language
  • Quality checklists by document type (Guides, API Docs, README, Architecture)
  • Validation patterns for problematic terms: ableist, gendered, violent language, time-based references, excessive claims, overused politeness
  • Writing style principles: tone, voice, techniques, things to avoid
  • Tool usage patterns: finding docs, validating content, understanding context, large-scale operations
  • Common issues and fixes table with diagnosis and solutions
  • Implementation approach with 6-step methodology
  • Anti-patterns guide including outdated information, too much detail, unclear jargon, inconsistent terminology, cultural insensitivity
  • Comprehensive documentation standards for clarity, accessibility, and inclusivity
  • Support for Read, Write, Edit, Glob, Grep and Task tools

Changed

• fetch-full-content plugin (v1.1.0.0): Enhanced configuration flexibility
  • Made folder argument optional with default cache directory
  • Restricted allowed-tools to Python only (removed other tool types)

---

Full Changelog: v1.9.0.0...v1.10.0.0

v1.9.0.0

Added

• cross-platform-notification plugin (v1.0.0.0): New plugin for native system notifications
  • Support for Windows, macOS, and Linux with platform-specific notification systems
  • Windows toast notifications via PowerShell
  • macOS notifications via terminal-notifier
  • Linux notifications via notify-send
  • Automatic fallback to console output when native notifications unavailable
  • Hooks into Claude Code events for automated notifications

Changed

• session-protocol plugin (v1.1.0.0):

  • Breaking: Removed notification hook and related notification infrastructure
  • Notification functionality separated into dedicated cross-platform-notification plugin for cleaner feature separation
  • Simplified to focus on session continuity management
  • Removed PreCompact hook
  • SessionStart hook remains for auto-detection of existing session protocols

• claude-code-capabilities plugin (v1.8.0.0):

  • Removed /list-skills slash command - Claude Code v2.1.0 added a built-in /skills command, making the custom command obsolete

---

Full Changelog: v1.8.1.0...v1.9.0.0