Trending AI Projects & Tools Research Report
Date: 2026-02-28 | Researcher: Ada (subagent)
1. Trending AI Agent Frameworks & Tools (Jan–Feb 2026)
🔥 OpenClaw — #1 on GitHub Trending
- Repo: openclaw/openclaw — 239k stars, 46k forks
- 175k stars gained this month alone. The dominant personal AI assistant platform.
- We're running this. The ecosystem is exploding around us.
obra/superpowers — Agentic Skills Framework (65k ⭐)
- Repo: obra/superpowers — 27k stars this month
- Complete software development workflow built on composable "skills." Starts with brainstorming/spec, then auto-generates implementation plans, dispatches subagents per task with two-stage review (spec compliance + code quality).
- Key skills: brainstorming, writing-plans, subagent-driven-development, TDD, git worktrees, code review.
- Works with Claude Code, Cursor, Codex, OpenCode.
- Relevance: The subagent-driven-development pattern mirrors our multi-agent architecture. Could inform how we structure coding tasks across agents.
gsd-build/get-shit-done — Context Engineering for Claude Code (22k ⭐)
- Repo: gsd-build/get-shit-done — 13k stars this month
- Meta-prompting and spec-driven development system. Solves "context rot" — quality degradation as context window fills.
- Supports Claude Code, OpenCode, Gemini CLI, and Codex.
npx get-shit-done-cc@latestto install.- Relevance: Could be useful for K2 (tech agent) when doing extended coding sessions. The context engineering approach is worth studying.
badlogic/pi-mono — AI Agent Toolkit (18k ⭐)
- Repo: badlogic/pi-mono — 15k stars this month
- Monorepo with: unified multi-provider LLM API, agent runtime with tool calling, coding agent CLI, Slack bot, TUI/web UI libraries, vLLM pod management.
- Relevance: The unified LLM API and agent runtime could be useful reference architecture. The Slack bot delegation pattern is interesting for multi-agent setups.
thedotmack/claude-mem — Session Memory for Claude Code (32k ⭐)
- Repo: thedotmack/claude-mem — 16k stars this month
- Auto-captures everything Claude does during coding sessions, compresses with AI, re-injects relevant context into future sessions.
- Relevance: We already have our own memory system (daily notes + MEMORY.md), but their compression approach could improve our memory efficiency.
KeygraphHQ/shannon — Autonomous AI Security Scanner (26k ⭐)
- Repo: KeygraphHQ/shannon — 22k stars this month
- Fully autonomous AI hacker. 96.15% success rate on XBOW Benchmark (hint-free, source-aware).
- Relevance: Could be useful for security auditing Sean's homelab infrastructure.
VectifyAI/PageIndex — Vectorless RAG (19k ⭐)
- Repo: VectifyAI/PageIndex — 9k stars this month
- Document indexing for reasoning-based RAG without vector databases.
- Relevance: Could be useful for document-heavy domains like Cora's real estate work.
virattt/dexter — Financial Research Agent (17k ⭐)
- Repo: virattt/dexter
- Autonomous agent for deep financial research.
- Relevance: Potentially useful pattern for Cora's market analysis needs.
google/langextract — Structured Info Extraction (34k ⭐)
- Repo: google/langextract — 10k stars this month
- Python library for extracting structured information from unstructured text with source grounding and visualization.
Other Notable
- tobi/qmd (11k ⭐) — Mini CLI search engine for local docs/knowledge bases. All local, tracking SOTA approaches.
- ThePrimeagen/99 (4k ⭐) — Neovim AI agent. "Done right."
- openai/skills (10k ⭐) — Official Skills Catalog for Codex.
- huggingface/skills (7k ⭐) — HuggingFace's skills framework.
2. OpenClaw Ecosystem Developments
Latest Release Highlights (Feb 27, 2026)
From the releases page:
- External Secrets Management — Full
openclaw secretsworkflow (audit, configure, apply, reload) with runtime snapshot activation and strict validation. - ACP/Thread-bound Agents — ACP agents are now first-class runtimes for thread sessions. This is directly relevant to our multi-agent Ada/K2/Cora/Winston setup.
- Agent Routing CLI —
openclaw agents bindings,openclaw agents bind,openclaw agents unbindfor account-scoped route management. Channel-only to account-scoped binding upgrades. - Codex WebSocket Transport — WebSocket-first by default for openai-codex (with SSE fallback).
- Android Nodes — Device capability, status, info commands, and
notifications.listsupport. - Lobster Workflow Shell — See below.
Key Fixes
- Telegram/DM allowlist runtime inheritance hardened
- Delivery queue retry backoff to prevent starvation
- Typing indicator cleanup across all channels
- Temp dir permissions fix for Linux umask issues
OpenClaw Ecosystem Repos
| Repo | Description | Stars |
|---|---|---|
| openclaw/openclaw | Core platform | 239k |
| openclaw/clawhub | Skill directory | 3.2k |
| openclaw/skills | Archived skill versions | 1.6k |
| openclaw/lobster | Workflow shell | 646 |
| openclaw/nix-openclaw | Nix packaging | 461 |
| openclaw/openclaw-ansible | Ansible deployment | 424 |
Lobster — Workflow Shell (New)
- Repo: openclaw/lobster
- OpenClaw-native workflow shell: typed (JSON-first) pipelines, jobs, and approval gates.
- Agents can invoke workflows in one step, saving tokens and improving determinism.
- Supports YAML workflow files with steps, env, conditions, and approval gates.
- Relevance: This is huge for our setup. Instead of Ada re-planning multi-step operations, she could invoke pre-defined Lobster workflows. E.g., an inbox triage workflow, a PR monitoring workflow, a deployment pipeline.
3. Multi-Agent Architectures & Patterns
NevaMind-AI/memU — Memory for 24/7 Agents (11k ⭐)
- Repo: NevaMind-AI/memU — 6k stars this month
- Memory framework specifically built for always-on proactive agents like OpenClaw.
- Treats memory like a file system: categories → memory items, with cross-references.
- Key capabilities: 24/7 proactive agent memory, user intention capture, cost-efficient context via caching.
- Reduces LLM token cost for keeping agents always online.
- Relevance: Directly applicable. Our current memory system (daily .md files + MEMORY.md) is manual. memU could automate memory extraction, organization, and proactive intent prediction across all our agents.
Subagent-Driven Development (Superpowers pattern)
- Dispatch fresh subagent per task → two-stage review (spec compliance, then code quality) → continue or fix
- Each subagent gets minimal context (just its task), avoiding context window bloat
- Relevance: We already use subagents. This pattern suggests a more structured review loop.
ACP (Agent Communication Protocol) — Now in OpenClaw
- Thread-bound ACP agents as first-class runtimes
- Spawn/send dispatch integration, lifecycle controls, startup reconciliation
- Relevance: This is the formalization of what we're doing with Ada coordinating K2/Cora/Winston. Worth exploring ACP bindings for cleaner inter-agent communication.
Patterns Observed Across Projects
- Skills as modular capabilities — OpenClaw skills, Superpowers skills, Codex skills, HuggingFace skills. The "skill" abstraction is winning.
- Spec-driven development — Write specs first, then let agents implement. GSD, Superpowers, and others converge on this.
- Context engineering > prompt engineering — Managing what goes INTO context matters more than how you phrase the prompt. GSD's core insight.
- Memory as infrastructure — memU, claude-mem, and our own system all treat memory as a first-class concern, not an afterthought.
- Workflow determinism — Lobster and similar tools trade LLM flexibility for deterministic execution when the plan is known.
4. MCP (Model Context Protocol) Servers
Official MCP Registry
- URL: registry.modelcontextprotocol.io
- GitHub has added an "MCP Registry" feature to their platform navigation, signaling mainstream adoption.
- The modelcontextprotocol/servers repo (80k ⭐) houses reference implementations.
MCP Servers Worth Exploring
We already have mcporter skill installed. Key servers to consider:
| Server | Use Case | Relevance |
|---|---|---|
| GitHub MCP | PR reviews, issue management | K2's dev workflows |
| Filesystem MCP | Structured file access | All agents |
| PostgreSQL/SQLite MCP | Database queries | Homelab monitoring |
| Brave Search MCP | Web search (when API key works) | Research tasks |
| Google Drive MCP | Document access | Cora's real estate docs |
| Slack MCP | Team communication | If Sean uses Slack |
| Home Assistant MCP | Smart home control | Winston's domain |
GitHub MCP Registry Integration
- GitHub now has MCP as a first-class platform feature ("MCP Registry" in nav alongside Copilot, Spark, Models).
- This signals MCP is becoming the standard tool integration protocol.
5. Agentic Coding Tools & Workflows
steipete/gogcli — Google Suite CLI (5k ⭐)
- Repo: steipete/gogcli
- Gmail, GCal, GDrive, GContacts from the CLI. By OpenClaw's creator (steipete).
- We already have the
gogskill installed.
steipete/summarize — Universal Summarizer (5k ⭐)
- Repo: steipete/summarize
- Point at any URL/YouTube/Podcast or file, get the gist. CLI + Chrome Extension.
microsoft/playwright-cli (4k ⭐)
- CLI for Playwright actions: record, generate code, inspect selectors, screenshots.
- Relevance: Could enhance our browser automation capabilities.
Coding Agent Landscape
| Tool | Approach | Notes |
|---|---|---|
| Claude Code | CLI agent with tools | What we use via OpenClaw's coding-agent skill |
| OpenAI Codex | WebSocket-first, skills-based | Now has WebSocket transport in OpenClaw |
| Superpowers | Skills + subagent orchestration | Plugin for Claude Code/Cursor/Codex |
| GSD | Context engineering + spec-driven | Works across Claude Code, Gemini, Codex |
| Pi Coding Agent | Monorepo toolkit | Unified LLM API + agent runtime |
| ThePrimeagen/99 | Neovim-native AI agent | For vim users |
Key Workflow Patterns
- Spec → Plan → Subagent execution → Review (Superpowers)
- Context engineering to prevent rot (GSD)
- Memory compression for session continuity (claude-mem)
- Typed workflow pipelines for determinism (Lobster)
- Skills as the unit of capability (OpenClaw, Codex, HuggingFace)
Recommendations for Our Setup
- Explore memU — Could dramatically improve our agents' memory efficiency and enable proactive behavior without manual MEMORY.md maintenance.
- Set up Lobster workflows — Pre-define common multi-step operations (inbox triage, deployment checks, PR monitoring) to save tokens.
- Leverage ACP bindings — The new agent routing CLI could formalize Ada↔K2/Cora/Winston communication.
- Consider Superpowers for K2 — When K2 handles coding tasks, the subagent-driven-development pattern with structured review would improve quality.
- Update OpenClaw — The latest release has significant improvements to secrets management, agent routing, and Android node support.
- Explore PageIndex — Vectorless RAG could be useful for Cora's real estate document analysis without needing a vector DB in the homelab.
Report generated from GitHub trending, OpenClaw releases, and project READMEs. Brave Search API was unavailable (subscription token invalid).