Daily AI Implementation Scout Council

Today's ranked top 20

#1

newbuild nowBoth runtimes30 / 35

vercel-labs/agent-browser repo

Browser automation CLI for AI agents written in Rust. The annotate flag overlays numbered labels on interactive elements in screenshots. Supports anti-bot wall bypass, parallel multi-session execution without cross-contamination, Electron app automation (VS Code, Slack, Figma, Notion, Spotify), and cloud browser backends in Vercel Sandbox or AWS Bedrock AgentCore. Created January 11, 2026. Over 30,000 GitHub stars.

What it does for you: AXION agents currently need platform APIs to interact with websites; browser automation fills the gap for sites that have no API or block scripted access. YY can now give any Axion agent a real browser with anti-bot stealth, covering GHL form interactions, MCL social media checks, and client portals that block headless requests. The cloud browser backend means scheduled agents can run without YY's machine being on.

In practice: Vercel Labs releases tend to be production-calibrated rather than proof-of-concept. The Rust implementation and 30,000-plus stars in under eight months signal serious community adoption. The AGENTS.md integration pattern and cloud backend options suggest this was designed for production agent pipelines, not just local hacking.

For: Both runtimes. Rust CLI callable from any agent; Vercel Sandbox and AWS Bedrock AgentCore cloud browser backends available for headless runs.

Security3

Quality5

Auditability4

Useful to you5

Useful to community5

Buildable now4

Hermes4

Verdict: build now. 30,000-plus stars, Vercel Labs production pedigree, Rust CLI runs in any agent via shell, solves a real automation gap in AXION where no-API-key site interaction is currently impossible. Cloud browser mode removes the local-machine dependency for scheduled agents.

Build #1 vercel-labs/agent-browser: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/vercel-labs/agent-browser. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · 30,000-plus GitHub stars as of August 2026; Rust implementation confirmed; Vercel Sandbox and AWS Bedrock AgentCore backends documented; annotate flag with numbered element overlays in release notes; created January 11, 2026.

#2

newbuild nowClaude (Axion)29 / 35

2389-research/claude-plugins plugin

28 production-tested plugins for Claude Code built by Harper Reed and the 2389 Research team. Covers test-driven development, multi-agent orchestration, iterative refinement, binary reverse engineering on ARM64 and x86_64 using radare2 and Ghidra, and structured decision frameworks. Tiered-delegation plugin has Sonnet plan a spec into sprints while Haiku executes and self-verifies, cutting costs approximately 64 percent. Each plugin installs in one command.

What it does for you: The tiered-delegation plugin alone cuts YY's Claude Code API costs by approximately 64 percent for long coding sessions by routing Haiku at execution tasks and reserving Sonnet for planning. The TDD and multi-agent orchestration plugins add structure to AXION build sessions. Harper Reed and 2389 Research use these in daily production, so they are calibrated to real workloads rather than demos.

In practice: Harper Reed is a credible voice in the developer tool space with a track record of practical, production-quality tooling. 28 plugins from a single shop with a 64 percent cost claim that is specific and verifiable is a stronger signal than the typical 'useful utilities' collection. The binary RE plugin is a standout for security and reverse engineering work.

For: Claude (Axion). Claude Code plugin system; one-command install per plugin via the 2389 plugin marketplace.

Security4

Quality4

Auditability4

Useful to you5

Useful to community4

Buildable now5

Hermes3

Verdict: build now. One-command install, no build step, battle-tested by the authors' own team. The 64 percent cost reduction from tiered delegation is immediately verifiable on any AXION coding session. Not currently enabled in YY's settings.json enabledPlugins list.

Install: visit https://2389-research.github.io/claude-plugins and run the one-command install for each plugin in Claude Code. Not currently in YY's settings.json enabledPlugins list. Review before enabling. Copy

Source · 28 plugins documented at skills.2389.ai; tiered delegation claims approximately 64 percent cost reduction (Sonnet plans, Haiku executes); binary-re plugin documents radare2, Ghidra, GDB, and QEMU support for ARM64, ARMv7, and x86_64; marketplace at 2389-research.github.io/claude-plugins.

#3

build nowBoth runtimes30 / 35

aws/agent-toolkit-for-aws plugin

Official AWS-supported collection of MCP servers, skills, and plugins that give AI agents direct access to AWS APIs, sandboxed script execution, and real-time AWS documentation search. Maintained by AWS engineering teams. Covers the main AWS service surface via MCP tool calls, meaning agents can query S3, Lambda, EC2, and other services without custom API wrappers.

What it does for you: AXION agents currently interact with AWS services only through hand-written boto3 calls or CLI commands. This toolkit gives YY's agents and Hermes a standardised, officially supported MCP interface to all major AWS services. The real-time documentation search tool eliminates the hallucinated API signatures that break agent-generated AWS code. Sandboxed script execution lets agents run AWS CLI commands safely.

In practice: Official AWS pedigree means long-term maintenance commitment, unlike community MCP servers that go stale. The combination of API access, sandboxed execution, and live docs search in one package is the right architecture for agents that need to both learn and act on AWS infrastructure.

For: Both runtimes. Official AWS-supported MCP servers installable in any MCP-compatible agent including Claude Code and Hermes.

Security4

Quality4

Auditability5

Useful to you4

Useful to community5

Buildable now4

Hermes4

Verdict: build now. Official AWS support, covers the full AWS service surface via MCP, sandboxed execution for safety, and real-time docs search to prevent hallucinated API calls. MCP config setup is the only hurdle and is well-documented. Not currently in YY's settings.json enabledPlugins list.

Install: follow the README at github.com/aws/agent-toolkit-for-aws to add the MCP servers to your Claude Code or Hermes MCP configuration, scoped to the least-privilege IAM role you want agents to use. Not currently in YY's settings.json enabledPlugins list. Review before enabling. Copy

Source · Official AWS-supported repository confirmed active August 2026 at github.com/aws/agent-toolkit-for-aws; 3 capability areas documented in README: AWS API access, sandboxed script execution, and real-time documentation search; maintained by AWS engineering with ongoing commits.

#4

test firstBoth runtimes29 / 35

rohitg00/agentmemory repo

Persistent memory for AI coding agents rated number one on real-world benchmarks. Apache 2.0, TypeScript. Ships 12 auto-capture hooks that compress every tool call, prompt, and stop into observations. Triple-stream retrieval using BM25, vector search, and knowledge graph reranked on-device at P50 latency under 20ms on a laptop. Hourly sweeps compress raw observations into semantic memories, merge duplicates, decay stale rows, and emit audit rows. Claims 95.2 percent retrieval R@5 and 92 percent token reduction. Around 13,000 GitHub stars as of May 2026.

What it does for you: Every AXION agent session currently starts from zero context: YY has to re-explain project decisions, architecture choices, and past errors each time. Agentmemory hooks into Claude Code and captures all of that automatically, retrieving relevant memories at the start of each turn. The 92 percent token reduction claim, if it holds on AXION's codebase, would significantly cut daily API costs.

In practice: The architecture is serious: triple-stream retrieval plus on-device reranking at sub-20ms latency is not a demo benchmark. The audit row trail for every deletion is the kind of detail that indicates production-calibrated thinking rather than research code. The 92 percent token reduction claim is extraordinary and warrants independent testing before relying on it.

For: Both runtimes. TypeScript memory layer hooks into Claude Code, Codex CLI, Cursor, Gemini CLI, and OpenCode; Hermes can read the markdown memory store directly.

Security4

Quality4

Auditability5

Useful to you5

Useful to community5

Buildable now3

Hermes3

Verdict: test first. Architecture is sound and the use case is the exact gap in AXION's current setup. Test first because the 92 percent token reduction claim is extraordinary and the 12 hooks need a validation pass on YY's Windows 11 Claude Code environment before calling it done.

Build #4 rohitg00/agentmemory: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/rohitg00/agentmemory. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · Apache 2.0 license confirmed; 13,000 GitHub stars as of May 2026; 95.2 percent retrieval R@5 benchmark claimed; 92 percent token reduction claimed; P50 retrieval latency under 20ms documented; 12 auto-capture hooks detailed in README; triple-stream BM25 plus vector plus knowledge graph architecture confirmed.

#5

newtest firstBoth runtimes25 / 35

jackwener/opencli repo

OpenCLI transforms any website, Electron app, or local binary into a standardised command-line interface using your logged-in Chrome session. AGENT.md integration lets Claude Code, Cursor, Codex CLI, Gemini CLI, and 20-plus other agents discover and execute CLI commands against sites like Twitter/X, Reddit, YouTube, 小红书, and more. 27,833 GitHub stars as of April 2026. AI agents operate these sites without API keys, using the user's own authenticated session.

What it does for you: YY's content operations on Twitter/X and LinkedIn currently require manual browser work or platform API keys that have rate limits and approval delays. OpenCLI gives AXION agents no-API-key access to those platforms via YY's existing Chrome sessions. For MCL content monitoring and posting workflows, this removes the API dependency entirely.

In practice: 27,833 stars for a tool this specific is unusually high and indicates the CLI-for-agents pattern resonated broadly. The AGENT.md convention is consistent with Claude Code's emerging agent discovery standards. The Chrome session dependency is a real constraint for fully headless scheduled use, but for interactive AXION workflows it is not a blocker.

For: Both runtimes. Universal CLI Hub; any agent calls CLI commands via AGENT.md integration; browser session runs locally in YY's Chrome.

Security3

Quality4

Auditability4

Useful to you4

Useful to community4

Buildable now4

Hermes2

Verdict: test first. High star count and clear MCL content workflow use case make this worth testing immediately. The Chrome session dependency limits use to YY's live machine, so verify which AXION scenarios are compatible before integrating into scheduled agents.

Build #5 jackwener/opencli: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/jackwener/opencli. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · 27,833 GitHub stars confirmed by SkillsLLM as of April 2026; AGENT.md integration documented; 20-plus supported platforms listed including Twitter/X, Reddit, YouTube; Electron app support (VS Code, Slack, Discord) documented; created and maintained by jackwener.

#6

newtest firstClaude (Axion)25 / 35

JCodesMore/ai-website-cloner-template skill

Claude Code skill that clones any website with one command. Takes screenshots via a browser extension, extracts colors and components, spawns parallel sub-agents each assigned one page section in its own branch, compares the result against the original, and fixes discrepancies. Also works with Cursor, Codex CLI, and Gemini CLI. 31,753 GitHub stars as of August 10, 2026. Created March 2026.

What it does for you: YY regularly builds landing pages and event sites (here.now sites, MCL event pages, client hubs) that take hours to build manually. This skill reduces a full page replica to one Claude Code command. The comparison-and-fix loop catches component mismatches before delivery, reducing the back-and-forth with YY.

In practice: 31,753 stars six months after creation is viral adoption, not slow growth. The multi-agent branch-per-component approach mirrors patterns YY already uses in AXION builds. The browser extension dependency and Windows 11 compatibility are the unknowns to resolve before relying on it in production.

For: Claude (Axion). Claude Code skill template; spawns parallel sub-agents each building one section in its own branch, then compares and fixes against the original screenshot.

Security3

Quality3

Auditability4

Useful to you4

Useful to community5

Buildable now4

Hermes2

Verdict: test first. High star count and direct MCL and client site use case. Need to verify that the browser extension and the parallel branch pattern work on YY's Windows 11 Chrome setup before calling it build-now.

Build #6 JCodesMore/ai-website-cloner-template: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/JCodesMore/ai-website-cloner-template. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · 31,753 GitHub stars as of August 10, 2026 (confirmed by SkillsLLM); created March 2026 by JCodesMore; parallel branch architecture documented in README; works with 12-plus coding agents listed.

#7

newtest firstBoth runtimes26 / 35

browser-act/skills plugin

Browser automation skills for AI agents with anti-bot wall bypass, CAPTCHA solving, TLS rotation, residential proxies, stealth fingerprinting, and stable session management for authenticated browsing. Supports parallel multi-task execution, independent multi-session operation, and isolated multi-account browsing. Two usage modes: cloud-managed execution or local browser control. Last updated August 12, 2026. 5,300 GitHub stars.

What it does for you: For AXION workflows that need to bypass anti-bot protections (client portals, social platform data, ticketing systems), browser-act provides a managed layer that handles the fingerprinting and proxy work rather than requiring YY to maintain his own stealth Chrome setup. Cloud mode means Hermes can trigger browser tasks without a local Chrome session running.

In practice: 5,300 stars is modest but the feature set is unusually complete for a young project: TLS rotation, CAPTCHA solving, and residential proxies together indicate this was built for production scale rather than toy demos. The August 12 update shows active maintenance. The cloud mode is the key differentiator over OpenCLI.

For: Both runtimes. Two modes: fully cloud-managed browser execution (no local Chrome needed), or local browser control via agent workflow; both modes callable from Claude Code or Hermes.

Security3

Quality4

Auditability4

Useful to you4

Useful to community4

Buildable now4

Hermes3

Verdict: test first. Cloud browser mode unlocks scheduled and Hermes-driven browser automation without local Chrome. Test first to evaluate the per-session cost structure and verify CAPTCHA solving works on the specific platforms AXION agents need.

Install: follow README at github.com/browser-act/skills to add the BrowserAct skills to your Claude Code or agent config; choose cloud-managed or local mode. Review before enabling. Copy

Source · 5,300 GitHub stars; updated August 12, 2026; TLS rotation, residential proxies, CAPTCHA solving, and stealth fingerprinting documented; two usage modes (cloud-managed and local) documented; multi-session and multi-account isolation confirmed.

#8

newtest firstStandalone tool27 / 35

Gemini 3.7 Flash plugin

Gemini 3.7 Flash launched August 13, 2026, three weeks after 3.6 Flash. Keeps 1M token input context and 64k output limit with multimodal input (text, image, video, audio, PDF), function calling, search as a tool, and computer use. Introductory pricing at 0.75 USD per million input tokens and 3.75 USD per million output tokens through December 31, 2026. Coding and agent benchmarks: DeepSWE v1.1 improved from 49.0 percent to 65.3 percent; AutomationBench from 17.0 to 30.4 percent; WebDev Arena Elo from 1538 to 1588.

What it does for you: At 0.75 USD per million input tokens, Gemini 3.7 Flash is a cost-effective backend for AXION tasks that need a capable model but not Claude's reasoning depth: content reformatting, first-pass summarisation, and bulk data extraction. The 65.3 percent DeepSWE score and WebDev Arena jump make it credible for coding subtasks in a tiered agent setup where Claude handles planning and Gemini handles cheaper execution steps.

In practice: Google released 3.7 Flash three weeks after 3.6 Flash, suggesting a very fast iteration cadence. The benchmark jumps are substantial for a Flash-tier model. The 0.75 USD introductory price is a competitive signal aimed at making it the default execution-tier model before the full price kicks in December 31.

For: Standalone tool. Google-hosted model API; accessed via the Gemini SDK or Google AI Studio; not a local tool.

Security4

Quality5

Auditability3

Useful to you4

Useful to community5

Buildable now4

Hermes2

Verdict: test first. Significant benchmark gains on coding and agent tasks at a low introductory price. Worth evaluating as the Haiku-equivalent execution-tier model in a Claude-plans/Gemini-executes tiered setup for AXION, but needs a head-to-head comparison against Claude Haiku on actual AXION tasks before committing.

Install: add the gemini-3.7-flash model to your agent via pip install google-generativeai and the Gemini API key from Google AI Studio at aistudio.google.com. Review before enabling. Copy

Source · DeepSWE v1.1 benchmark 49.0 to 65.3 percent; AutomationBench 17.0 to 30.4 percent; WebDev Arena Elo 1538 to 1588; launch date August 13, 2026 confirmed; introductory price 0.75 USD per million input tokens through December 31, 2026; 1M token context and computer use confirmed.

#9

newtest firstClaude (Axion)25 / 35

steipete/claude-code-mcp plugin

MCP server that runs Claude Code in one-shot mode, turning any Claude Code session into a spawner of sub-sessions. Permissions are bypassed by default, with an optional native Claude Code permission mode. Enables true recursive agent delegation: a top-level Claude Code session can hand off a subtask to a fresh Claude Code instance that has its own file access, tool calls, and context.

What it does for you: AXION orchestrator sessions currently dispatch sub-agents at the API level via the Workflow tool. Steipete's MCP lets YY run agent-in-agent trees natively inside a single Claude Code conversation, with sub-sessions inheriting project context automatically. This is the simplest path to recursive Claude Code delegation without writing a custom orchestration layer.

In practice: Peter Steinberger (steipete) has a long track record of production-quality iOS and Mac tools; his open-source work is usually well-engineered. The default permission bypass is the principal risk: any sub-session can write or delete files without prompting. The optional native permission mode addresses this, but it must be explicitly enabled.

For: Claude (Axion). MCP server that runs inside Claude Code and exposes Claude Code itself as a callable tool, enabling one Claude Code session to delegate subtasks to fresh Claude Code sub-sessions.

Security2

Quality4

Auditability4

Useful to you5

Useful to community4

Buildable now3

Hermes3

Verdict: test first. Powerful capability for AXION multi-agent patterns with minimal setup. Test first because the default permission bypass needs to be evaluated against AXION's file security requirements before enabling in any session that touches production folders.

Install: add steipete/claude-code-mcp to your MCP config in Claude Code settings and explicitly set the permission mode to native before use. Review before enabling. Copy

Source · GitHub repository confirmed live at steipete/claude-code-mcp; one-shot mode documented; permission bypass described as default in README; optional native Claude Code permission mode documented; steipete has 7,000-plus GitHub followers.

#10

test firstBoth runtimes27 / 35

airblackbox/gateway repo

Open-source EU AI Act compliance scanner and trust layer. 51 checks across EU AI Act Articles 9 through 15. Drop-in integration for LangChain, CrewAI, AutoGen, and OpenAI agents. Local-first with no data leaving the machine. Includes ConsentGate for risk-classifying tool calls and InjectionDetector for scanning prompts. The EU AI Act compliance deadline for deploying AI systems in the EU was August 2, 2026, which has now passed.

What it does for you: YY's enterprise clients (IFPAS, MCL) face EU AI Act obligations if they deploy AI systems to EU-based audiences. Airblackbox provides the compliance audit trail and risk classifications those deployments require. Adding it as a gateway layer to AXION agents takes the posture from undocumented to auditable without rewriting the underlying agents.

In practice: The August 2, 2026 EU AI Act deadline passed 14 days ago, making this immediately relevant rather than forward-looking. 51 checks across specific Articles 9 to 15 is concrete and auditable. Local-first processing is the right call for a compliance tool: sending compliance data to a third-party cloud would itself create a compliance concern.

For: Both runtimes. Drop-in middleware layer for LangChain, CrewAI, AutoGen, and OpenAI agents; also wraps direct SDK calls; language-agnostic compliance checks run locally.

Security5

Quality4

Auditability4

Useful to you4

Useful to community4

Buildable now3

Hermes3

Verdict: test first. EU AI Act obligations are live as of August 2. Drop-in integration means no agent rewrite needed, just a gateway wrapper. Test first because the 51 checks need a pass against each AXION agent's actual tool set to identify which risk categories apply before committing to the full configuration.

Build #10 airblackbox/gateway: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/airblackbox/gateway. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · 51 checks across EU AI Act Articles 9 to 15 documented; local-first execution with no external data transmission confirmed; drop-in integrations for LangChain, CrewAI, AutoGen, and OpenAI documented; EU AI Act compliance deadline August 2, 2026 confirmed; Show HN post on Hacker News at item 47141347.

#11

newwatchBoth runtimes26 / 35

microsoft/skills repo

Official Microsoft repository of skills, MCP servers, custom agents, and Agents.md files for grounding coding agents. Includes mcp-builder (guides creation of high-quality MCP servers), skill-creator (packaging guide for agent skills), and Microsoft Foundry plugin integration. Copilot code review with agent skills and MCP went generally available July 29, 2026 for Copilot Pro, Pro+, Business, and Enterprise users.

What it does for you: The mcp-builder and skill-creator patterns in this repo are directly applicable to YY's AXION skill-building workflow. The Agents.md convention for grounding agents with live API documentation reduces hallucinated AWS, Azure, and SDK patterns in agent-generated code. Even without a GitHub Copilot plan, the patterns are portable to Claude Code via CLAUDE.md.

In practice: Microsoft's official skills repo going live alongside Copilot code review MCP GA is a signal that the SKILL.md and AGENTS.md conventions are becoming a de facto standard across coding agent platforms. The patterns here will likely appear in Anthropic's own recommendations within one to two quarters.

For: Both runtimes. Skills, MCP servers, and AGENTS.md files usable in Claude Code, GitHub Copilot, and any SDK-compatible coding agent.

Security4

Quality4

Auditability5

Useful to you3

Useful to community4

Buildable now3

Hermes3

Verdict: watch. The patterns are useful for YY's skill-building workflow, but most of the value requires a GitHub Copilot plan that YY does not have. Watch and extract the Agents.md and SKILL.md conventions for use in CLAUDE.md grounding without needing the full Copilot integration.

Build #11 microsoft/skills: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/microsoft/skills. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · Copilot code review agent skills and MCP GA announced July 29, 2026 (GitHub Changelog); mcp-builder and skill-creator subdirectories confirmed in repository; Agents.md file at microsoft/skills/blob/main/Agents.md confirmed; microsoft.github.io/skills/ live documentation site.

#12

newwatchStandalone tool24 / 35

GitHub Copilot SDK for Java repo

Client library that lets server-side Java applications create Copilot agent sessions, register tools, send prompts, and receive structured responses. Works with Jakarta EE and Spring via CompletableFuture, annotations, lambdas, and virtual threads. BYOK: bring your own API key from OpenAI, Anthropic, Azure, AWS Bedrock, Google AI Studio, or any OpenAI-compatible provider, with usage billed to that provider rather than GitHub. Fine-grained tool access control via ToolSet. Agent Plugins 1.0 now generally available in VS Code, Copilot CLI, and the Copilot app.

What it does for you: The BYOK architecture is the notable pattern here: it shows how to embed a coding agent runtime into a server-side application while letting the customer supply their own API keys and billing. If YY ever productises AXION features for enterprise clients (IFPAS, Ryan's coaching), this SDK's ToolSet permission model is the reference architecture for scoping what the embedded agent can access.

In practice: The BYOK model is strategically significant: it turns Copilot into a runtime layer rather than a locked-in SaaS, which enterprise buyers prefer. The Java-specific details are not directly relevant to YY's Python and Node stack, but the ToolSet and agent session patterns translate to any language's SDK.

For: Standalone tool. Java server-side SDK; not directly usable in YY's Python and Node stack but architectural patterns are portable.

Security4

Quality4

Auditability5

Useful to you2

Useful to community4

Buildable now3

Hermes2

Verdict: watch. Java is not YY's primary stack, so direct adoption is low-priority. The BYOK and ToolSet patterns are worth understanding as architecture references for productising AXION agent capabilities for enterprise clients.

Build #12 GitHub Copilot SDK for Java: use the ai-implementation-build-intake skill to build this safely. Source: https://github.blog/engineering/using-the-github-copilot-sdk-for-java/. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · GitHub blog post August 10, 2026 confirmed; BYOK supporting OpenAI, Anthropic, Azure, AWS Bedrock, and Google AI Studio documented; Agent Plugins 1.0 GA in VS Code, Copilot CLI, and Copilot app confirmed; fine-grained ToolSet control documented; Jakarta EE and Spring support confirmed.

#13

newwatchClaude (Axion)23 / 35

composio-community/awesome-claude-plugins repo

Community-curated list of Claude Code plugins and MCP servers, organised by category. Maintained by the Composio community team. Covers plugins for file operations, shell commands, browser automation, code analysis, and MCP server creation. Includes an mcp-builder plugin for creating high-quality MCP servers that integrate external APIs with Claude Code.

What it does for you: When YY is looking for a plugin to handle a specific AXION task, this list is a faster starting point than searching GitHub cold. The mcp-builder entry is directly useful for any new MCP server YY needs to write for AXION services.

In practice: Composio is a credible player in the agent tools space. A community-maintained plugin list by a named organisation is more reliable than random awesome-lists with no clear maintainer. The categorisation makes it scannable quickly.

For: Claude (Axion). Curated reference list of Claude Code plugins; not itself a runnable tool.

Security4

Quality3

Auditability4

Useful to you3

Useful to community4

Buildable now3

Hermes2

Verdict: watch. Reference list rather than a runnable tool; bookmark for discovery when looking for new Claude Code plugins rather than building with it directly.

Build #13 composio-community/awesome-claude-plugins: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/composio-community/awesome-claude-plugins. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · GitHub repository confirmed at composio-community/awesome-claude-plugins as of August 2026; mcp-builder plugin entry confirmed; Composio has been active in the agent tools space since 2023; categorised plugin listing with at least 12 entries across 4 categories confirmed.

#14

newwatchBoth runtimes25 / 35

Sagar-Gogineni/agentguard repo

EU AI Act compliance middleware for AI agents. Makes any LLM-powered agent legally deployable in the EU with three lines of code. Wraps the LLM API call to enforce risk classification, consent gates, and audit logging per EU AI Act requirements. Deadline for compliance was August 2, 2026.

What it does for you: Lighter-weight alternative to airblackbox/gateway if YY needs a quick compliance wrapper for a single agent rather than the full 51-check scan. The three-lines-of-code promise means a compliance layer can be added to any AXION agent in minutes to demonstrate due diligence to enterprise clients.

In practice: An individual-contributor repository with a bold three-lines-of-code claim for EU AI Act compliance. The claim needs scrutiny: EU AI Act compliance is not reducible to three lines in a real audit. This is more likely a useful first-pass risk classification and logging layer than a complete compliance solution.

For: Both runtimes. Python middleware that wraps any LLM API call; three lines of code to integrate.

Security5

Quality3

Auditability4

Useful to you3

Useful to community4

Buildable now4

Hermes2

Verdict: watch. Three-lines-of-code claim is appealing for quick integration but the scope of coverage compared to airblackbox's 51-check approach is unclear. Watch alongside airblackbox to compare depth before choosing which approach to adopt.

Build #14 Sagar-Gogineni/agentguard: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/Sagar-Gogineni/agentguard. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · GitHub repository confirmed at Sagar-Gogineni/agentguard; EU AI Act compliance deadline August 2, 2026 confirmed; three-line integration claim stated in README; Python middleware targeting any LLM API call confirmed.

#15

newwatchBoth runtimes23 / 35

xyTom/coding-tools-mcp plugin

MCP server that gives AI agents the ability to code. Supports the MCP 2026-07-28 specification in full. Tools is the only advertised capability surface; exposes coding tools to any agent via the standardised MCP protocol. Runs locally against the codebase the agent is working on.

What it does for you: For AXION agents that need to code across multiple sessions or within an agent-in-agent pattern, this MCP server provides a clean tool surface conformant with the latest MCP spec. Useful as a reference implementation if YY needs to build a custom coding tool MCP for AXION.

In practice: Minimal information in public search results beyond the MCP 2026-07-28 spec compliance claim. The spec compliance is the main value signal: the July 28 MCP spec is the current production standard, so a server built against it will not need immediate updates.

For: Both runtimes. MCP server conformant with the MCP 2026-07-28 specification; installable in any MCP-compatible agent.

Security4

Quality3

Auditability4

Useful to you3

Useful to community3

Buildable now3

Hermes3

Verdict: watch. Limited public information; needs more investigation before adopting. MCP 2026-07-28 compliance is a positive signal. Watch until more community feedback surfaces.

Install: follow README at github.com/xyTom/coding-tools-mcp to add the MCP server to your Claude Code MCP configuration. Review before enabling. Copy

Source · GitHub repository at xyTom/coding-tools-mcp confirmed; MCP 2026-07-28 specification support documented; tools-only capability surface stated in README.

#16

newwatchClaude (Axion)24 / 35

auchenberg/claude-code-mcp plugin

Full implementation of Claude Code as an MCP server. Provides tools for file operations, shell commands, and code analysis, making a complete Claude Code environment accessible as a set of MCP tool calls from any parent agent. Alternative to steipete/claude-code-mcp with a different implementation approach: this one exposes the full Claude Code surface rather than running it in one-shot mode.

What it does for you: Provides another path to agent-in-agent Claude Code delegation, complementing steipete's one-shot approach. Useful for comparing which implementation fits AXION's orchestration patterns better before committing to one.

In practice: Kenneth Auchenberg (auchenberg) is a former Microsoft and Stripe developer tools team lead; this is credible pedigree for developer tool work. The full surface exposure versus one-shot mode is a meaningful architectural difference from steipete's implementation.

For: Claude (Axion). Full Claude Code as an MCP server; exposes file operations, shell commands, and code analysis as MCP tools callable from another agent.

Security3

Quality3

Auditability4

Useful to you4

Useful to community4

Buildable now3

Hermes3

Verdict: watch. Evaluate alongside steipete/claude-code-mcp to choose the better fit for AXION's orchestration patterns before implementing either.

Install: add auchenberg/claude-code-mcp to your MCP configuration in Claude Code settings. Compare with steipete/claude-code-mcp before choosing which to enable. Review before enabling. Copy

Source · GitHub repository at auchenberg/claude-code-mcp confirmed active August 2026; full Claude Code implementation as MCP server documented; 3 tool categories confirmed: file operations, shell commands, and code analysis; author auchenberg has 10-plus years in developer tooling including roles at Microsoft and Stripe.

#17

watchBoth runtimes29 / 35

github/github-mcp-server plugin

Official GitHub MCP server maintained by GitHub engineering. Recent additions include the get_commits method and the get_file_blame tool for attribution queries. Covers pull request creation, issue management, file operations, search, and repository management via MCP tool calls. Improved rate limit error messages help agents handle GitHub API throttling gracefully.

What it does for you: AXION agents that interact with the axion-skills GitHub remote, the docs-tracked git repo, or any client GitHub repo benefit from the new get_file_blame tool for attribution queries and the improved rate limit messaging that prevents silent failures in overnight runs.

In practice: This is an update to a foundational MCP server that most GitHub-connected agents already use. The get_file_blame tool is a small but useful addition for code review agents. Rated here as a re-list because the additions are material for AXION's overnight agent runs.

For: Both runtimes. Official GitHub MCP server; installable in any MCP-compatible agent.

Security4

Quality4

Auditability5

Useful to you4

Useful to community5

Buildable now4

Hermes3

Verdict: watch. Already a baseline tool for most GitHub-connected agents. Re-listed because get_commits and get_file_blame are specifically useful for AXION's code review and attribution workflows.

Install: follow the README at github.com/github/github-mcp-server to update to the latest version in your MCP configuration and enable the new get_commits and get_file_blame tools. Review before enabling. Copy

Source · get_commits method and get_file_blame tool confirmed in recent release notes as of August 2026; improved rate limit error messages documented; official GitHub engineering maintenance confirmed; 2 new tools in this update cycle: get_commits and get_file_blame; releases page at github.com/github/github-mcp-server/releases active.

#18

newwatchStandalone tool20 / 35

multica-ai/multica repo

Multi-platform AI agent interface with version v0.4.26 released August 14, 2026. This release adds inbox keyboard shortcuts, DeepSeek Harness support for lower-cost agent runs, and agent task improvements. Positions as a local-first 24/7 agent coworker compatible with OpenClaw, Hermes, Claude Code, Codex, and OpenCode across 20-plus CLI agents.

What it does for you: The DeepSeek Harness support in v0.4.26 is the interesting detail: if multica can route agent tasks through DeepSeek at lower cost while maintaining task quality, it is a cost-reduction complement to the tiered delegation pattern from 2389-research/claude-plugins. Worth monitoring to see if the task routing quality holds.

In practice: Small project with a specific niche (unified UI for multiple CLI agents). v0.4.26 on August 14 indicates active development. DeepSeek Harness support is a forward-looking cost optimisation signal. Too early to assess community traction.

For: Standalone tool. Multi-platform AI agent interface (desktop application); not designed to be embedded in another agent pipeline.

Security3

Quality3

Auditability3

Useful to you3

Useful to community3

Buildable now3

Hermes2

Verdict: watch. DeepSeek Harness support is a cost-reduction angle worth monitoring, but the project is still early. Watch for community feedback on task routing quality before investing setup time.

Build #18 multica-ai/multica: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/multica-ai/multica. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · v0.4.26 released August 14, 2026 (releases page confirmed); DeepSeek Harness support documented in release notes; inbox keyboard shortcuts and agent task improvements listed; compatibility with Claude Code, Codex, and 20-plus CLI agents stated.

#19

newwatchBoth runtimes24 / 35

ark-forge/mcp-eu-ai-act plugin

Open-source MCP server that detects EU AI Act violations in codebases. Free tier requires no sign-up or API key. Pro version unlocks the full compliance toolkit. Complements static analysis tools with EU AI Act-specific checks. Released near the August 2, 2026 EU AI Act compliance deadline.

What it does for you: Adds EU AI Act codebase scanning as an MCP tool call in any Claude Code or Hermes session. For YY's enterprise client builds, running this MCP on the agent code before delivery provides a documented compliance check without manual legal review of each file.

In practice: MCP-native approach to compliance scanning is architecturally clean: it fits into the existing agent workflow rather than requiring a separate CI step. The free tier with no sign-up is a low-friction entry point. Coverage depth relative to airblackbox's 51-check approach is unknown.

For: Both runtimes. MCP server that scans codebases for EU AI Act violations; callable from any MCP-compatible agent.

Security5

Quality3

Auditability3

Useful to you3

Useful to community4

Buildable now3

Hermes3

Verdict: watch. MCP-native compliance scanning is a useful pattern. Watch to compare coverage with airblackbox/gateway before choosing which compliance layer to standardise on.

Install: add ark-forge/mcp-eu-ai-act to your MCP configuration; free tier requires no sign-up. Compare coverage with airblackbox/gateway before choosing. Review before enabling. Copy

Source · GitHub repository at ark-forge/mcp-eu-ai-act confirmed; free tier with no sign-up documented; MCP server targeting EU AI Act violations in codebases confirmed; pro version mentioned for full compliance toolkit; released near August 2, 2026 deadline.

#20

newwatchBoth runtimes23 / 35

databricks-solutions/ai-dev-kit repo

Databricks Toolkit for Coding Agents provided by Databricks Field Engineering. Gives AI agents direct access to Databricks workspace APIs, notebook execution, SQL warehouse queries, and cluster management via MCP tool calls. Designed for teams building agent workflows on top of Databricks infrastructure.

What it does for you: If any AXION data pipeline or MCL analytics work touches Databricks or Delta Lake, this toolkit gives YY's agents native Databricks access without writing custom API wrappers. Field Engineering backing means it reflects real production patterns rather than a community prototype.

In practice: Databricks Field Engineering repos tend to be practical and production-tested, created for customer deployments rather than demos. The Databricks ecosystem is large and YY's direct use is currently low, but as MCL analytics matures this becomes more relevant.

For: Both runtimes. MCP servers and skills for coding agents; usable with Claude Code, Cursor, and other coding agents via standard MCP configuration.

Security4

Quality3

Auditability4

Useful to you2

Useful to community4

Buildable now3

Hermes3

Verdict: watch. Not immediately relevant to YY's current stack, but Databricks is a common enterprise analytics platform. Watch for when client work brings Databricks into scope.

Build #20 databricks-solutions/ai-dev-kit: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/databricks-solutions/ai-dev-kit. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · GitHub repository at databricks-solutions/ai-dev-kit confirmed active August 2026; Field Engineering maintenance label confirmed; 4 capability areas documented in README: Databricks workspace APIs, notebook execution, SQL warehouse queries, and cluster management; Databricks serves over 10,000 enterprise customers globally.