@techwavedev/agi-agent-kit 1.3.4 → 1.3.6

package/CHANGELOG.md

@@ -5,362 +5,114 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
-## [1.3.4] - 2026-02-16
+## [1.3.6] - 2026-02-16
+
+### Security
+
+- **Backend Template Hardening** — Resolved 5 high-severity vulnerabilities in `loki-mode` backend example by upgrading dependencies.
+
+### Added
+
+- **OpenClaw Platform Support** — Complete integration with auto-detection and instruction symlinks.
+  - **Badge**: Added OpenClaw badge to README
+  - **Ecosystem**: Now supporting 9 major AI coding platforms
+
+- **Hybrid BM25+Vector Memory Search** — True hybrid retrieval combining Qdrant vector similarity with SQLite FTS5 keyword search:
+  - **BM25 Index** (`bm25_index.py`): SQLite FTS5 sidecar for exact keyword matching (error codes, IDs, env vars)
+  - **Weighted Score Merge**: `finalScore = 0.7 × vectorScore + 0.3 × textScore` (configurable)
+  - **3 Search Modes**: `hybrid` (default), `vector`, `keyword`
+  - **Auto-indexing**: Every `store_memory()` call automatically indexes into BM25
+  - **`bm25-sync` command**: Rebuild keyword index from existing Qdrant collection
+  - **Graceful fallback**: Falls back to vector-only if BM25 unavailable
+  - **FTS5 query sanitization**: Handles special characters (hyphens, dots) in search terms
+  - **16/16 tests passing**: 4 new BM25 tests (index, hybrid, score merge, fallback)
+
+## [1.3.5] - 2026-02-16
 
 ### Added
 
-- **Extended Skills Tier: 862 Total Skills** — Integrated 782 community skills from [antigravity-awesome-skills](https://github.com/sickn33/antigravity-awesome-skills) (v5.4.0) by [@sickn33](https://github.com/sickn33). Restructured skill tiers:
+- **Extended Skills Tier: 861 Total Skills** — Integrated 782 community skills from [antigravity-awesome-skills](https://github.com/sickn33/antigravity-awesome-skills) (v5.4.0). Restructured skill tiers:
   - **Core** (4 skills): webcrawler, pdf-reader, qdrant-memory, documentation
-  - **Medium** (75 skills): Core + knowledge-tier specialized skills across 16 categories
+  - **Medium** (75 skills): Core + specialized skills across 16 categories
   - **Full** (861 skills): Medium + 782 community skills, all adapted for AGI framework
-  - CLI installer renamed: `knowledge` → `medium`, added `full` tier
 
-- **Categorized Skill Organization** — All skills (knowledge + extended) physically organized into 16 domain categories:
-  - `frontend/`, `backend/`, `ai-agents/`, `devops/`, `testing/`, `security/`, `architecture/`, `mobile/`, `debugging/`, `documentation/`, `workflow/`, `content/`, `data/`, `gaming/`, `blockchain/`, `other/`
-  - Core tier stays flat (4 skills). Knowledge: 75 skills in 16 categories. Extended: 782 skills in 16 categories.
-  - `category_map.json` — Persistent mapping for upstream sync; new skills default to `other/` for later reclassification.
-  - Installer (`init.js`) auto-detects categories vs skills via `SKILL.md` presence; installs flat at user destination.
-  - `SKILLS_CATALOG.md` — Browsable catalog with tier badges (🟢 Core, 🔵 Medium, 🟡 Full) and quick-reference table.
+- **Categorized Skill Organization** — All skills organized into 16 domain categories:
+  `frontend/`, `backend/`, `ai-agents/`, `devops/`, `testing/`, `security/`, `architecture/`, `mobile/`, `debugging/`, `documentation/`, `workflow/`, `content/`, `data/`, `gaming/`, `blockchain/`, `other/`
 
-- **8-Platform Support** — Expanded from 2 to 8 platform symlinks:
-  - Instruction files: `GEMINI.md`, `CLAUDE.md`, `OPENCODE.md`, `COPILOT.md` (→ `AGENTS.md`)
-  - Skill directories: `.claude/skills/`, `.gemini/skills/`, `.codex/skills/`, `.cursor/skills/`, `.adal/skills/` (→ `skills/`)
-  - Compatible with: Claude Code, Gemini CLI, Codex CLI, Antigravity IDE, Cursor, GitHub Copilot, OpenCode, AdaL CLI
+- **8-Platform Support** — Claude Code, Gemini CLI, Codex CLI, Antigravity IDE, Cursor, GitHub Copilot, OpenCode, AdaL CLI
 
-- **AGI Framework Adaptation** — All 782 community skills automatically adapted with:
+- **AGI Framework Adaptation** — All 782 community skills adapted with:
   - Qdrant Memory Integration (semantic caching)
   - Agent Team Collaboration (orchestrator-driven invocation)
   - Local LLM Support (Ollama embeddings)
-  - Source attribution linking back to original repo
 
-- **Attribution & Licensing** — Proper open-source compliance:
-  - `THIRD-PARTY-LICENSES.md` — Full MIT license text + upstream attribution chain
-  - `templates/skills/extended/ATTRIBUTION.md` — Credits with support links
-  - `templates/skills/extended/LICENSE-MIT` — Original upstream license file
-
-- **Automated Adaptation Pipeline** — CI/CD for upstream sync:
-  - `scripts/adapt_extended_skills.py` — Diff detection, section-marker preservation, categorized placement, dry-run mode
-  - `.github/workflows/adapt-skills.yml` — Auto-creates PR on upstream changes
-
-### Changed
-
-- **6 Enriched Skills** — Merged richer external versions into knowledge tier:
-  - `code-review-checklist` (+9KB), `mcp-builder` (+5.7KB), `seo-fundamentals` (+3KB)
-  - `systematic-debugging` (+2.8KB), `test-driven-development` (+2.7KB), `testing-patterns` (+2.3KB)
-- **Installer** (`bin/init.js`) — Updated `copySkills()` to walk category subdirectories; detects skill vs category by `SKILL.md` presence; installs all skills flat at destination.
-- **Adaptation Script** (`scripts/adapt_extended_skills.py`) — Now places skills into correct category subdirectories using `category_map.json`; `find_skill_path()` walks nested dirs; auto-updates map with new upstream skills.
-- `README.md` — Updated for v1.3.3: new tier descriptions, 8-platform table, categorized structure, community credits section
-- `package.json` — Version bump to 1.3.3, added `THIRD-PARTY-LICENSES.md` to published files
+- **Automated NPM Publishing** — GitHub Actions workflow: create a release → NPM publishes automatically via OIDC Trusted Publisher
 
 ### Credits
 
-This release would not be possible without the [Antigravity Awesome Skills](https://github.com/sickn33/antigravity-awesome-skills) community. Skills aggregated from 50+ contributors including Anthropic, Microsoft, Vercel Labs, Supabase, Trail of Bits, Expo, Sentry, and many more. See [SOURCES.md](https://github.com/sickn33/antigravity-awesome-skills/blob/main/docs/SOURCES.md) for the full attribution ledger.
+This release includes skills aggregated from 50+ open-source contributors including Anthropic, Microsoft, Vercel Labs, Supabase, Trail of Bits, and many more. See [SOURCES.md](https://github.com/sickn33/antigravity-awesome-skills/blob/main/docs/SOURCES.md) for the full attribution ledger.
 
 ## [1.2.8] - 2026-02-14
 
 ### Added
 
-- **Superpowers Adaptation**: Adapted best patterns from [obra/superpowers](https://github.com/obra/superpowers) into the agi framework, extending it with structured execution, TDD enforcement, and verification gates while preserving all existing multi-platform, memory, and agent capabilities.
-
-  **New Skills:**
-  - **`executing-plans`**: Structured plan execution with two modes — Batch (3 tasks + human checkpoint) or Subagent-Driven (fresh agent per task + two-stage review: spec compliance then code quality). Platform-adaptive across Claude Code, Gemini, Kiro, and Opencode.
-  - **`test-driven-development`**: Iron-law TDD enforcement with RED-GREEN-REFACTOR cycle, rationalization prevention table, and verification checklist. "No production code without a failing test first."
-  - **`verification-before-completion`**: Universal evidence gate — no completion claims without fresh verification output. Integrates with agi audit scripts (`checklist.py`, `security_scan.py`, `lint_runner.py`, etc.).
-
-### Changed
-
-- **`systematic-debugging`**: Replaced lightweight 110-line version with comprehensive 4-phase methodology (Root Cause Investigation → Pattern Analysis → Hypothesis Testing → Implementation). Includes iron law, multi-component evidence gathering, rationalization table, and real-world impact stats (95% vs 40% first-time fix rate).
-- **`plan-writing`**: Added TDD step structure template (write test → verify fail → implement → verify pass → commit), bite-sized task granularity (2-5 min per step), and execution handoff to `executing-plans` skill.
-- **`brainstorming`**: Added HARD-GATE (no code before design approval), propose 2-3 approaches step, and design document saving with handoff to `plan-writing`.
-- **`orchestrator`** agent: Added two-stage review protocol (spec compliance then code quality), execution mode selection (batch vs subagent), verification gate, and referenced skills table.
-- **`parallel-agents`**: Added focused agent prompt structure template, explicit "when NOT to use" heuristic, and review-and-integrate protocol with `verification-before-completion` gate.
-- **`debugger`** agent: Added enhanced skills integration table referencing `systematic-debugging`, `test-driven-development`, and `verification-before-completion`.
-- **`SKILLS_CATALOG.md`**: Regenerated with 45 template skills (3 new: `executing-plans`, `test-driven-development`, `verification-before-completion`).
-
-### Why This Is More Complete Than Superpowers
-
-> All superpowers patterns were adopted. The agi framework extends them with capabilities superpowers does not have.
-
-| Capability                    |  obra/superpowers   |                  agi Framework                  |
-| ----------------------------- | :-----------------: | :---------------------------------------------: |
-| TDD Enforcement               |         ✅          |                  ✅ (adapted)                   |
-| Plan Execution with Review    |         ✅          |        ✅ (adapted + platform-adaptive)         |
-| Systematic Debugging          |         ✅          |   ✅ (adapted + `debugger` agent integration)   |
-| Verification Gates            |         ✅          |      ✅ (adapted + agi script integration)      |
-| Two-Stage Code Review         |         ✅          |         ✅ (adapted into orchestrator)          |
-| Git Worktrees                 |         ✅          |           ➖ (standard branches used)           |
-| Multi-Platform Orchestration  | ❌ Claude Code only | ✅ 4 platforms (Claude, Gemini, Kiro, Opencode) |
-| Semantic Memory (Qdrant)      |         ❌          |            ✅ 90-100% token savings             |
-| 19 Specialist Agents          |         ❌          |          ✅ Domain-specific boundaries          |
-| Agent Boundary Enforcement    |         ❌          |         ✅ File-type ownership protocol         |
-| Dynamic Question Generation   |         ❌          |         ✅ Trade-off tables, priorities         |
-| 12 Audit/Verification Scripts |         ❌          |     ✅ Security, lint, UX, SEO, Lighthouse      |
-| Memory-First Protocol         |         ❌          |          ✅ Auto cache-hit before work          |
-| Skill Creator + Catalog       |         ❌          |            ✅ 45+ composable skills             |
-| Platform Setup Wizard         |         ❌          |         ✅ One-shot auto-configuration          |
+- **Superpowers Adaptation** — Adapted best patterns from [obra/superpowers](https://github.com/obra/superpowers):
+  - **`executing-plans`**: Structured plan execution with Batch or Subagent-Driven modes + two-stage review
+  - **`test-driven-development`**: Iron-law TDD enforcement with RED-GREEN-REFACTOR cycle
+  - **`verification-before-completion`**: No completion claims without fresh verification output
+  - **`systematic-debugging`**: Comprehensive 4-phase methodology (Root Cause → Pattern Analysis → Hypothesis → Implementation)
 
 ## [1.2.7] - 2026-02-11
 
 ### Security & Maintenance
 
-- **Repository Sanitization**: Removed all internal development configurations (`.agent`, `.claude`, `.gemini`) and private skills from the public repository.
-- **Workflow Hardening**: Enforced fork-and-PR workflow in `CONTRIBUTING.md`.
-- **Community Standards**: Added issue templates and PR templates.
+- Repository sanitization: removed all internal development configurations from public repository
+- Workflow hardening: enforced fork-and-PR workflow
+- Added issue templates and PR templates
 
 ## [1.2.6] - 2026-02-10
 
 ### Added
 
-- **notebooklm-rag (Deep RAG)**: MCP-first autonomous knowledge backend powered by Google NotebookLM + Gemini. The agent fully manages NotebookLM via MCP tools — authentication, library CRUD, querying with auto follow-ups, and Qdrant caching for token savings and cross-session context keeping. Includes fallback Python scripts (Patchright browser automation) when MCP is unavailable. Opt-in for users with Google accounts; default RAG remains `qdrant-memory`. Based on [PleasePrompto/notebooklm-skill](https://github.com/PleasePrompto/notebooklm-skill) (MIT).
-
-### Removed
-
-- **notebooklm-mcp**: Merged into `notebooklm-rag`. The comprehensive Deep RAG skill supersedes the basic MCP connector.
-
-### Fixed
-
-- **npmignore**: Fixed deep-path glob patterns (`**/__pycache__/`, `**/*.pyc`, `**/data/`) preventing compiled Python files and sensitive data from leaking into the published NPM package.
-
-## [1.2.5] - 2026-02-09
-
-### Fixed
-
-- **Documentation Updates**: README updated to reflect current version and removed internal skill references.
-
-## [1.2.4] - 2026-02-09
-
-### Fixed
-
-- **Minor bug fixes**: Addressed issues from v1.2.3 release.
+- **NotebookLM RAG Skill**: Deep-search RAG powered by Google NotebookLM + Gemini. MCP-first autonomous knowledge backend with Qdrant caching for token savings.
 
 ## [1.2.3] - 2026-02-09
 
 ### Added
 
-- **Auto Python Virtual Environment**: `npx init` now auto-creates `.venv/` and installs all dependencies — eliminates the `externally-managed-environment` error on macOS.
-- **Auto Platform Setup**: `npx init` runs `platform_setup.py --auto` after venv creation, pre-configuring platform-specific settings (`.claude/settings.json`, `.claude/skills/`) and showing remaining manual steps.
-- **Activation Reference Table** (README): Comprehensive reference with 4 sections — 14 slash commands, 8 `@agent` mentions, 18 natural language trigger keyword categories, and 6 memory system commands.
-- **Semantic Memory Section** (README): Setup guide with Qdrant + Ollama commands, token savings table (90-100% savings), and usage examples.
-- **`pyright`** added to `requirements.txt` for Python type checking out of the box.
-
-### Changed
-
-- **`requirements.txt`**: Expanded from 6 to 12 packages across 5 organized sections (Core, Memory, Embeddings, Cloud, Testing & Auditing). Every dependency is documented with inline comments.
-- **`bin/init.js`**: Added `setupPythonEnv()` and `runPlatformSetup()` functions. Uses `child_process.execSync` for venv creation and `pip install`. Cross-platform support (macOS/Linux/Windows paths).
-- **README Prerequisites**: No longer tells users to `pip install` manually — now references auto-created `.venv` with activation command.
-- **Post-install message**: Step 1 is now "Activate the Python environment" instead of "Install Python dependencies."
-- **Templates synced**: `requirements.txt`, `README.md` copied to `templates/base/` for NPX distribution.
-- **Documentation SKILL.md**: `Last Updated` timestamp refreshed to 2026-02-09.
-- **SKILLS_CATALOG.md**: Regenerated with 56 skills.
-
-### Fixed
-
-- **`ModuleNotFoundError: No module named 'yaml'`**: `pyyaml` was missing from `requirements.txt` — `system_checkup.py` crashed on fresh installs. Now included in core dependencies.
-- **Missing dependencies**: `gitpython`, `ollama`, `sentence-transformers`, `playwright` were used by skills but not listed in `requirements.txt`. All now included.
-- **`--auto-apply` flag**: `init.js` called `platform_setup.py` with non-existent `--auto-apply` flag — fixed to use existing `--auto` flag.
-
-## [1.2.2] - 2026-02-09
-
-### Added
-
-- **Platform-Adaptive Multi-Agent Orchestration** (`parallel-agents` v2.0):
-  - **Strategy A: Claude Code Agent Teams** — True parallel multi-agent orchestration via tmux/in-process sessions. Requires `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1`.
-  - **Strategy B: Claude Code Subagents** — Background/foreground task orchestration via `Task()` tool with `context: fork`.
-  - **Strategy C: Sequential Personas** — Universal fallback for Gemini, Opencode, and other platforms using `@agent` persona switching.
-  - **Strategy D: Kiro Autonomous Agent** — Async parallel task execution in sandboxed environments with PR-based delivery and cross-repo coordination.
-  - Automatic platform detection and strategy selection.
-  - Subagent configuration reference (frontmatter, tools, permissions, model, memory).
-  - Full Claude Code subagent lifecycle documentation (create, configure, invoke, manage).
-
-- **Kiro IDE Powers & Autonomous Agent Support**:
-  - Full Kiro Powers documentation: `POWER.md` anatomy, frontmatter keywords activation, onboarding sections, steering files, `mcp.json` configuration with auto-namespacing.
-  - Kiro Hooks system (`.kiro/hooks/`) with JSON structure and automated behaviors.
-  - Kiro Autonomous Agent: sandboxed execution, PR-based reviews, cross-repo coordination, learning from code reviews.
-  - 12 curated launch partner Powers documented (Figma, Supabase, Stripe, Neon, Netlify, Postman, Strands, Datadog, Dynatrace, AWS CDK, Terraform, Aurora).
-  - **Antigravity ↔ Kiro mapping guide**: Skill-to-Power conversion workflow, directory mapping, frontmatter translation.
-
-- **Plugin & Extension Auto-Discovery** (`plugin-discovery` v1.1.0 — NEW SKILL):
-  - Cross-platform extension auto-discovery for Claude Code, Kiro IDE, Gemini, and Opencode.
-  - Claude Code: Plugin marketplace guide (official + custom), LSP plugins by language, service integrations, scopes, subagent/skill discovery.
-  - Kiro: Full Powers discovery, installation guide (IDE, kiro.dev, GitHub, local path), setup checklist.
-  - Gemini/Opencode: Skills catalog discovery, MCP server detection.
-  - Cross-platform compatibility map (11 features × 4 platforms).
-
-- **One-Shot Platform Setup Wizard** (`platform_setup.py`):
-  - Auto-detects platform: Claude Code, Kiro IDE, Gemini, Opencode.
-  - Scans project tech stack: languages (JS/TS/Python/Go/Rust/Ruby), frameworks (Next.js/React/Vue/Express/Angular/Svelte/Astro/React Native), services (GitHub/GitLab/Docker/Vercel/Netlify/Stripe/Supabase/Firebase/Terraform).
-  - Generates prioritized recommendations with `🔴 High` / `🟡 Medium` / `🟢 Low` indicators.
-  - Auto-applies configurable settings (Agent Teams, directory creation, hook setup) with single `Y/n` confirmation.
-  - Shows manual instructions for platform-only actions (plugin installs, Power installs).
-  - 4 modes: interactive, `--auto`, `--dry-run`, `--json`.
-  - New `/setup` workflow with `// turbo` annotation for auto-run.
-
-- **Intelligent Routing v2.0** (`intelligent-routing` v2.0):
-  - Platform detection at session start (Claude Code, Kiro IDE, Gemini, Opencode).
-  - Proactive capability announcements per platform.
-  - Team Leader mode for Claude Code with Agent Teams enabled.
-  - Powers-driven orchestration mode for Kiro IDE.
-  - Proactive feature recommendations: suggests Agent Teams, plugins, Powers, Autonomous Agent when not enabled.
-  - Multi-domain task routing adapts to best available parallelism strategy.
-
-- **Memory System Integration** (`session_boot.py` — NEW SCRIPT):
-  - `execution/session_boot.py` — Single entry point for session initialization. Checks Qdrant, Ollama, embedding models, and collections in one command. `--auto-fix` flag auto-pulls missing models and creates collections.
-  - **Session Boot Protocol** added as the **first section** in `AGENTS.md` — agents run `python3 execution/session_boot.py --auto-fix` before any work begins.
-  - `AGENTS.md` Memory-First section rewritten with explicit CLI commands: `memory_manager.py auto`, `store`, `cache-store` with decision tree table.
-  - `platform_setup.py` now detects full memory system: Qdrant status, Ollama status, embedding model presence, collection existence, point counts.
-  - Memory recommendations engine: suggests starting Qdrant/Ollama, pulling embedding model, or initializing collections when issues are detected.
-  - Memory report section (🧠) in platform setup output shows live system status.
-
-### Changed
-
-- **`parallel-agents`**: Rewritten from v1.0 to v2.0 — now platform-adaptive with 4 orchestration strategies instead of 1.
-- **`intelligent-routing`**: Rewritten from v1.0 to v2.0 — now includes platform detection, proactive recommendations, and Kiro support.
-- **Cross-Platform Compatibility Map**: Updated with 2 new rows (Dynamic MCP Loading, Cross-Repo Tasks) and accurate Kiro feature coverage.
-- **`package.json`**: Added `kiro`, `opencode`, and `platform-adaptive` keywords for NPM discoverability.
-- **Templates**: All modified skills (`parallel-agents`, `intelligent-routing`, `plugin-discovery`) and scripts (`platform_setup.py`) synced to `templates/skills/knowledge/` for NPX distribution.
-- **SKILLS_CATALOG.md**: Regenerated with 56 skills, including the new `plugin-discovery` skill.
-- **`bin/init.js`**: NPX init now copies `execution/` scripts (session_boot.py, session_init.py, memory_manager.py) and `directives/` (memory_integration.md) to scaffolded projects. Update function also refreshes these files.
-- **Post-install message**: Now shows `session_boot.py --auto-fix` as step 3 after installation.
-- **`/setup-memory` workflow**: Fixed stale references, now uses correct scripts (session_init.py, memory_manager.py).
-- **`/setup` workflow**: Added memory system troubleshooting section with Docker/Ollama/collection init commands.
-
-### Fixed
-
-- **Stale script references**: `/setup-memory` workflow referenced non-existent `init_memory_system.py` and `memory_middleware.py` — now correctly uses `session_init.py` and `memory_manager.py`.
-- **Memory system unused**: Despite being installed, no agent instructions explicitly told agents WHEN to call the memory scripts. Now enforced via Session Boot Protocol at top of AGENTS.md.
-- **`platform_setup.py` missing memory**: Setup wizard reported platform features but ignored Qdrant/Ollama status — now detects and recommends fixes.
-- **`info` action missing**: Platform setup wizard didn't handle informational recommendations (e.g., "collections are empty") — added `info` action type with ℹ️ status icon.
-- **Templates missing execution scripts**: NPX init created the `execution/` directory but didn't copy any scripts into it — now ships session_boot.py, session_init.py, and memory_manager.py.
+- **Auto Python Virtual Environment**: `npx init` now auto-creates `.venv/` and installs all dependencies
+- **Auto Platform Setup**: `npx init` runs `platform_setup.py --auto` after venv creation
+- **Activation Reference Table**: 14 slash commands, 8 `@agent` mentions, 18 trigger keyword categories, 6 memory commands
+- **Platform-Adaptive Multi-Agent Orchestration**: 4 strategies (Agent Teams, Subagents, Sequential Personas, Kiro Autonomous Agent)
+- **Plugin & Extension Auto-Discovery**: Cross-platform extension discovery for Claude Code, Kiro IDE, Gemini, and Opencode
+- **One-Shot Platform Setup Wizard**: Auto-detects platform, scans project stack, generates recommendations, auto-applies settings
+- **Memory System Integration**: `session_boot.py` — single entry point for Qdrant + Ollama initialization
 
 ## [1.1.7] - 2026-02-07
 
 ### Added
 
-- **NotebookLM RAG Skill** (EC Scoped): New `notebooklm-rag` skill providing deep-search RAG capabilities powered by Google NotebookLM + Gemini 2.5. Complementary to `qdrant-memory` for explicit, document-grounded research.
-  - **B.L.A.S.T. Protocol**: Structured research workflow (Browse → Load → Ask → Synthesize → Transfer).
-  - **4 Research Modes**: Quick (single query), Deep (3-5 iterative), Cross-Ref (multi-notebook), Plan (doc-grounded planning).
-  - **Confidence Classification**: Automatic HIGH/MEDIUM/LOW categorization with source attribution.
-  - `scripts/research_query.py` — Generates structured research questions by mode.
-  - `scripts/research_report.py` — Formats findings into Markdown/JSON reports with confidence levels and knowledge gap analysis.
-  - `scripts/preflight_check.py` — Pre-flight validation checklist for MCP server, auth, and library.
-  - `references/research_patterns.md` — Workflow templates, query optimization, rate limit strategies.
-- **Auto-Update in Skill Creator**: `init_skill.py` now auto-runs `update_catalog.py` and `sync_docs.py` after creating a new skill.
-  - Default ON — use `--no-auto-update` to skip.
-  - New `--skills-dir` flag for specifying catalog update target directory.
-  - Non-fatal: graceful warnings if documentation skill is not installed.
-
-### Changed
-
-- **NotebookLM MCP Skill**: Migrated to `PleasePrompto/notebooklm-mcp` browser-automated implementation with full auth, library management, and stealth mode support.
-- **Skill Creator Documentation** (`SKILL_skillcreator.md`):
-  - Step 3: Documents new auto-update behavior and `--no-auto-update` / `--skills-dir` flags.
-  - Step 6: Notes that manual catalog update is now only needed for modifications/deletions.
-- **SKILLS_CATALOG.md**: Regenerated with 55 skills including `notebooklm-rag`.
-
-### Fixed
-
-- **`init_skill.py` argument parsing**: Migrated from fragile positional parsing (`sys.argv`) to robust `argparse` with proper help text and validation.
-- **`sync_docs.py` invocation**: Fixed `--update-catalog` flag to pass `"true"` as required value (not just the bare flag).
-
-## [1.1.6] - 2026-02-07
-
-### Added
-
-- **37 Knowledge Skills Promoted to Root**: All knowledge-pack skills now live at `skills/` root alongside core skills for unified access:
-  - `api-patterns`, `app-builder`, `architecture`, `bash-linux`, `behavioral-modes`, `brainstorming`, `clean-code`, `code-review-checklist`, `database-design`, `deployment-procedures`, `documentation-templates`, `frontend-design`, `game-development`, `geo-fundamentals`, `i18n-localization`, `intelligent-routing`, `lint-and-validate`, `mcp-builder`, `mobile-design`, `nextjs-best-practices`, `nodejs-best-practices`, `parallel-agents`, `performance-profiling`, `plan-writing`, `powershell-windows`, `python-patterns`, `react-patterns`, `red-team-tactics`, `seo-fundamentals`, `server-management`, `systematic-debugging`, `tailwind-patterns`, `tdd-workflow`, `testing-patterns`, `vulnerability-scanner`, `webapp-testing`.
-- **NotebookLM MCP Skill**: New `notebooklm-mcp` skill to interact with Google NotebookLM via Model Context Protocol, with documentation, example script, and reference assets.
-- **Opencode Support**: Added `OPENCODE.md` configuration file for Opencode editor compatibility with `opencode-antigravity-auth` plugin.
-- **Execution Scripts**: Two new deterministic execution scripts:
-  - `memory_manager.py` — Centralized memory management for Qdrant operations.
-  - `session_init.py` — Session initialization and context bootstrapping.
-- **Stitch Skills in Templates**: `design-md`, `react-components`, and `stitch-loop` now included in knowledge templates (`templates/skills/knowledge/`) for NPX distribution.
-- **Self-Update in Templates**: `self-update` skill with `update_kit.py` added to knowledge templates.
-- **Knowledge SKILLS_CATALOG**: New comprehensive `SKILLS_CATALOG.md` inside `templates/skills/knowledge/` for template-level skill discovery.
-
-### Changed
-
-- **AGENTS.md**: Added "Getting Started" section with installation, dependency, and update instructions.
-- **Memory Integration Directive**: Complete rewrite of `directives/memory_integration.md` — simplified from 211 lines to 95 lines with clearer goal/inputs/execution structure and Ollama embedding documentation.
-- **Skill Creator**: Updated `SKILL_skillcreator.md` with refined skill creation guidelines.
-- **SKILLS_CATALOG.md**: Expanded with 560+ lines of new skill entries covering all promoted knowledge skills.
-- **Documentation Skill**: Updated `skills/documentation/SKILL.md` with improved detection and sync logic.
-- **Template Hierarchy Restructured**: Shifted standalone template skills (`design-md`, `react-components`, `stitch-loop`) into `templates/skills/knowledge/` for consistent NPX packaging.
-
-### Improved
-
-- **Core Skill Definitions**: Refined SKILL.md files across `pdf-reader`, `qdrant-memory`, and `webcrawler` with updated descriptions, triggers, and references.
-- **Qdrant Memory Scripts**: Updated all 7 scripts (`benchmark_token_savings.py`, `embedding_utils.py`, `hybrid_search.py`, `init_collection.py`, `memory_retrieval.py`, `semantic_cache.py`, `test_skill.py`) with improved patterns.
-- **Knowledge Skill Documentation**: Mass update of SKILL.md definitions across 36+ knowledge skills with enhanced frontmatter, triggers, and reference content in templates.
-- **Audit Scripts**: Updated `ux_audit.py`, `security_scan.py`, `lighthouse_audit.py`, `test_runner.py`, `playwright_runner.py`, and `seo_checker.py` in templates.
-
-### Fixed
-
-- **Version Bump**: Package version correctly set to `1.1.6` in `package.json`.
-- **Template Cleanup**: Removed duplicate standalone Stitch skill templates (`templates/skills/design-md/`, `templates/skills/react-components/`, `templates/skills/stitch-loop/`), consolidated into `templates/skills/knowledge/`.
-- **Removed `verify_public_release.py`**: Deprecated top-level verification script removed (functionality integrated into CI/CD pipeline).
+- **NotebookLM MCP Skill**: Interact with Google NotebookLM via Model Context Protocol
+- **37 Knowledge Skills Promoted**: All knowledge-pack skills now live at `skills/` root for unified access
+- **Opencode Support**: Added `OPENCODE.md` configuration
 
 ## [1.1.5] - 2026-01-27
 
 ### Added
 
-- **Stitch Skills Suite**: Added three new skills from Google Stitch:
-  - `design-md`: Analyzes Stitch projects to synthesize semantic `DESIGN.md` systems.
-  - `react-components`: Converts Stitch screens into modular, validating React components.
-  - `stitch-loop`: Orchestrates autonomous iterative website building loops.
-- **Memory Integration**: Integrated `qdrant-memory` into the Stitch skills suite for:
-  - Storing design systems for future retrieval (`design-md`).
-  - Retrieving code patterns and interfaces (`react-components`).
-  - Leveraging past decisions and project context (`stitch-loop`).
-
-## [1.1.2] - 2026-01-23
-
-### Added
-
-- **Self-Update Skill**: New `self-update` skill with `update_kit.py` script for easy framework updates.
-- **System Checkup**: New `system_checkup.py` execution script to verify agents, skills, workflows, and scripts.
-- **Workflows**: Added `/checkup` and `/update` workflows for quick access.
-
-### Changed
-
-- Updated `README.md` with comprehensive Quick Start, Commands, and Architecture sections.
-- Updated `SKILLS_CATALOG.md` to include self-update skill.
-
-## [1.1.01] - 2026-01-23
-
-### Fixed
-
-- Stabilized Full Suite activation.
-- Micro-bump for polish releases.
+- **Stitch Skills Suite**: `design-md`, `react-components`, `stitch-loop` with Qdrant memory integration
 
 ## [1.1.0] - 2026-01-23
 
 ### Added
 
-- **Knowledge Pack**: Imported 36+ new skills from `agents-web` including:
-  - `api-patterns`: Best practices for REST/GraphQL/tRPC.
-  - `frontend-design`: UX/UI principles and audit tools.
-  - `security-auditor`: Vulnerability scanning and red-team tactics.
-  - `mobile-design`: iOS/Android development patterns.
-- **Agent Roles**: Added `project-planner`, `orchestrator`, and `security-auditor` agent personas.
-- **Rules**: Added global `deployment_policy` and `clean-code` standards.
-
-### Changed
-
-- Refactored `init` command to support `knowledge` pack.
-- Enhanced `AGENTS.md` with new agent capabilities.
-- Updated `SKILLS_CATALOG.md` with full list of new skills.
-
-## [1.0.1] - 2026-01-23
-
-### Fixed
-
-- **Security**: Removed all references to private infrastructure from public templates.
-- **Safety**: Added `verify_public_release.py` to prevent accidental publication of private secrets.
-- **Menu**: Fixed `init` menu showing internal options.
+- **Knowledge Pack**: 36+ new skills from `agents-web` (API patterns, frontend design, security, mobile)
+- **Agent Roles**: `project-planner`, `orchestrator`, `security-auditor`
+- **Clean-code** and **deployment** global standards
 
 ## [0.1.0] - 2026-01-23
 
 ### Initial Release
 
-- Core framework with `webcrawler`, `pdf-reader`, and `qdrant-memory`.
-- CLI tool `agi-agent-kit` for scaffolding.
+- Core framework with `webcrawler`, `pdf-reader`, and `qdrant-memory`
+- CLI tool `agi-agent-kit` for scaffolding

package/README.md

@@ -1,13 +1,38 @@
-# AGI Agent Kit
+# 🚀 AGI Agent Kit
 
-**Enterprise-Grade Agentic Framework & Scaffolding Tool**
+> **Stop hallucinating. Start executing.**
 
 [![npm version](https://img.shields.io/npm/v/@techwavedev/agi-agent-kit.svg)](https://www.npmjs.com/package/@techwavedev/agi-agent-kit)
+[![npm downloads](https://img.shields.io/npm/dw/@techwavedev/agi-agent-kit.svg)](https://www.npmjs.com/package/@techwavedev/agi-agent-kit)
 [![License: Apache-2.0](https://img.shields.io/badge/License-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
+[![Claude Code](https://img.shields.io/badge/Claude%20Code-Anthropic-purple)](https://claude.ai)
+[![Gemini CLI](https://img.shields.io/badge/Gemini%20CLI-Google-blue)](https://github.com/google-gemini/gemini-cli)
+[![Codex CLI](https://img.shields.io/badge/Codex%20CLI-OpenAI-green)](https://github.com/openai/codex)
+[![Cursor](https://img.shields.io/badge/Cursor-AI%20IDE-orange)](https://cursor.sh)
+[![GitHub Copilot](https://img.shields.io/badge/GitHub%20Copilot-VSCode-lightblue)](https://github.com/features/copilot)
+[![OpenCode](https://img.shields.io/badge/OpenCode-CLI-gray)](https://github.com/opencode-ai/opencode)
+[![Antigravity IDE](https://img.shields.io/badge/Antigravity-DeepMind-red)](https://github.com/techwavedev/agi-agent-kit)
+[![AdaL CLI](https://img.shields.io/badge/AdaL%20CLI-SylphAI-pink)](https://sylph.ai/)
+[![OpenClaw](https://img.shields.io/badge/OpenClaw-CLI-teal)](https://github.com/openclaw/openclaw)
+[![Buy me a coffee](https://img.shields.io/badge/Buy%20me%20a-coffee-d13610?logo=buymeacoffee&logoColor=white)](https://www.buymeacoffee.com/eltonmachado)
 
-`@techwavedev/agi-agent-kit` is a modular, deterministic framework designed to bridge the gap between LLM reasoning and reliable production execution. It scaffolds a "3-Layer Architecture" (Intent → Orchestration → Execution) that forces agents to use tested scripts rather than hallucinating code.
+**AGI Agent Kit** is the enterprise-grade scaffolding that turns any AI coding assistant into a **deterministic production machine**. While LLMs are probabilistic (90% accuracy per step = 59% over 5 steps), this framework forces them through a **3-Layer Architecture** — Intent → Orchestration → Execution — where business logic lives in tested scripts, not hallucinated code.
 
-**v1.3.4** — Now with **861 curated skills** (including 782 community skills from [antigravity-awesome-skills](https://github.com/sickn33/antigravity-awesome-skills)), **8-platform support** (Claude Code, Gemini CLI, Codex CLI, Cursor, Copilot, OpenCode, AdaL CLI, Antigravity IDE), structured plan execution, TDD enforcement, verification gates (adapted from [obra/superpowers](https://github.com/obra/superpowers)), and semantic memory.
+### Why this exists
+
+Most AI coding setups give you a prompt and hope for the best. AGI Agent Kit gives you:
+
+- 🧠 **Hybrid Memory** — Qdrant vectors + BM25 keywords: semantic similarity for concepts, exact matching for error codes and IDs (90-100% token savings)
+- 🎯 **19 Specialist Agents** — Domain-bounded experts (Frontend, Backend, Security, Mobile, Game Dev...) with enforced file ownership
+- ⚡ **861 Curated Skills** — 4 core + 75 professional + 782 community skills across 16 domain categories
+- 🔒 **Verification Gates** — No task completes without evidence. TDD enforcement. Two-stage code review.
+- 🌐 **9 Platforms, One Config** — Write once, run on Claude Code, Gemini CLI, Codex CLI, Cursor, Copilot, OpenCode, AdaL CLI, Antigravity IDE, OpenClaw
+
+```bash
+npx @techwavedev/agi-agent-kit init
+```
+
+If this project helps you, consider [supporting it here](https://www.buymeacoffee.com/eltonmachado) or simply ⭐ the repo.
 
 ---
 
@@ -54,7 +79,7 @@ This checks Qdrant, Ollama, embedding models, and collections — auto-fixing an
 | **Verification Gates**        | Evidence before claims — no completion without fresh verification output                      |
 | **Platform-Adaptive**         | Auto-detects Claude Code, Gemini CLI, Codex CLI, Cursor, Copilot, OpenCode, AdaL, Antigravity |
 | **Multi-Agent Orchestration** | Agent Teams, subagents, Powers, or sequential personas — adapts to platform                   |
-| **Semantic Memory**           | Built-in Qdrant-powered memory with 95% token savings via caching                             |
+| **Hybrid Memory**             | Qdrant vectors + BM25 keywords with weighted score merge (95% token savings)                  |
 | **Self-Healing Workflows**    | Agents read error logs, patch scripts, and update directives automatically                    |
 | **One-Shot Setup**            | Platform detection + project stack scan + auto-configuration in one command                   |
 
@@ -154,81 +179,6 @@ python3 execution/memory_manager.py auto \
 # → {"source": "cache", "cache_hit": true, "tokens_saved_estimate": 12}
 ```
 
----
-
-## 🧪 Real Benchmark: Subagents vs Agent Teams
-
-The framework supports two orchestration modes. Here are **real test results** from `execution/benchmark_modes.py` running on local infrastructure (Qdrant + Ollama `nomic-embed-text`, zero cloud API calls):
-
-```
-MODE A: SUBAGENTS — Independent, fire-and-forget
-  📤 Explore Auth Patterns    → ✅ stored in cache + memory (127ms)
-  📤 Query Performance        → ❌ FAILED (timeout — fault tolerant)
-  📤 Scan CVEs                → ✅ stored in cache + memory (14ms)
-  Summary: 2/3 completed, 1 failed, 0 cross-references
-
-MODE B: AGENT TEAMS — Shared context, coordinated
-  👤 Backend Specialist       → ✅ stored in shared memory (14ms)
-  👤 Database Specialist      → ✅ stored in shared memory (13ms)
-  👤 Frontend Specialist      → 🔗 Read Backend + Database output first
-     ✅ Got context from team-backend: "API contract: POST /api/messages..."
-     ✅ Got context from team-database: "Schema: users(id UUID PK, name..."
-     → ✅ stored in shared memory (14ms)
-  Summary: 3/3 completed, 0 failed, 2 cross-references
-```
-
-**2nd run (cache warm):** All queries hit cache at **score 1.000**, reducing total time from 314ms → 76ms (Subagents) and 292ms → 130ms (Agent Teams).
-
-| Metric               | Subagents                            | Agent Teams                          |
-| -------------------- | ------------------------------------ | ------------------------------------ |
-| Execution model      | Fire-and-forget (isolated)           | Shared context (coordinated)         |
-| Tasks completed      | 2/3 (fault tolerant)                 | 3/3                                  |
-| Cross-references     | 0 (not supported)                    | 2 (peers read each other's work)     |
-| Context sharing      | ❌ Each agent isolated               | ✅ Peer-to-peer via Qdrant           |
-| Two-stage review     | ❌                                   | ✅ Spec + Quality                    |
-| Cache hits (2nd run) | 5/5                                  | 5/5                                  |
-| Embedding provider   | Ollama local (nomic-embed-text 137M) | Ollama local (nomic-embed-text 137M) |
-
-**Try it yourself:**
-
-```bash
-# 1. Start infrastructure
-docker run -d -p 6333:6333 -v qdrant_storage:/qdrant/storage qdrant/qdrant
-ollama serve & ollama pull nomic-embed-text
-
-# 2. Boot memory system
-python3 execution/session_boot.py --auto-fix
-# ✅ Memory system ready — 5 memories, 1 cached responses
-
-# 3. Run the full benchmark (both modes)
-python3 execution/benchmark_modes.py --verbose
-
-# 4. Or test individual operations:
-
-# Store a decision (embedding generated locally via Ollama)
-python3 execution/memory_manager.py store \
-  --content "Chose PostgreSQL for relational data" \
-  --type decision --project myapp
-# → {"status": "stored", "point_id": "...", "token_count": 5}
-
-# Auto-query: checks cache first, then retrieves context
-python3 execution/memory_manager.py auto \
-  --query "what database did we choose?"
-# → {"source": "memory", "cache_hit": false, "context_chunks": [...]}
-
-# Cache an LLM response for future reuse
-python3 execution/memory_manager.py cache-store \
-  --query "how to set up auth?" \
-  --response "Use JWT with 24h expiry, refresh tokens in httpOnly cookies"
-
-# Re-query → instant cache hit (score 1.000, zero re-computation)
-python3 execution/memory_manager.py auto \
-  --query "how to set up auth?"
-# → {"source": "cache", "cache_hit": true, "tokens_saved_estimate": 12}
-```
-
----
-
 ## 🌐 Platform Support
 
 The framework automatically detects your AI coding environment and activates the best available features.
@@ -312,15 +262,16 @@ The system operates on three layers:
 
 ---
 
-## 🧠 Semantic Memory
+## 🧠 Hybrid Memory (BM25 + Vector)
 
-Built-in Qdrant-powered memory with automatic token savings:
+Dual-engine retrieval: Qdrant vector similarity for semantic concepts + SQLite FTS5 BM25 for exact keyword matching. Automatically merges results with configurable weights.
 
 | Scenario              | Without Memory | With Memory | Savings  |
 | --------------------- | -------------- | ----------- | -------- |
 | Repeated question     | ~2000 tokens   | 0 tokens    | **100%** |
 | Similar architecture  | ~5000 tokens   | ~500 tokens | **90%**  |
 | Past error resolution | ~3000 tokens   | ~300 tokens | **90%**  |
+| Exact ID/code lookup  | ~3000 tokens   | ~200 tokens | **93%**  |
 
 **Setup** (requires [Qdrant](https://qdrant.tech/) + [Ollama](https://ollama.com/)):
 
@@ -342,11 +293,27 @@ Agents automatically run `session_boot.py` at session start (first instruction i
 # Auto-query (check cache + retrieve context)
 python3 execution/memory_manager.py auto --query "your task summary"
 
-# Store a decision
+# Store a decision (auto-indexes into BM25)
 python3 execution/memory_manager.py store --content "what was decided" --type decision
 
-# Health check
+# Health check (includes BM25 index status)
 python3 execution/memory_manager.py health
+
+# Rebuild BM25 index from existing Qdrant data
+python3 execution/memory_manager.py bm25-sync
+```
+
+**Hybrid search modes** (via `hybrid_search.py`):
+
+```bash
+# True hybrid (default): vector + BM25 merged
+python3 scripts/hybrid_search.py --query "ImagePullBackOff error" --mode hybrid
+
+# Vector only (pure semantic)
+python3 scripts/hybrid_search.py --query "database architecture" --mode vector
+
+# Keyword only (exact BM25 match)
+python3 scripts/hybrid_search.py --query "sg-018f20ea63e82eeb5" --mode keyword
 ```
 
 ---

package/bin/init.js

@@ -336,6 +336,7 @@ function createSymlinks(targetPath) {
     { name: "CLAUDE.md", target: "AGENTS.md" },
     { name: "OPENCODE.md", target: "AGENTS.md" },
     { name: "COPILOT.md", target: "AGENTS.md" },
+    { name: "OPENCLAW.md", target: "AGENTS.md" },
   ];
 
   for (const link of instructionSymlinks) {
@@ -365,6 +366,7 @@ function createSymlinks(targetPath) {
     { platform: ".codex/skills", platformName: "Codex CLI" },
     { platform: ".cursor/skills", platformName: "Cursor" },
     { platform: ".adal/skills", platformName: "AdaL CLI" },
+    { platform: ".openclaw/skills", platformName: "OpenClaw" },
   ];
 
   for (const { platform, platformName } of platformDirs) {