Daily AI Implementation Scout Council

Today's ranked top 20

#1

test firstBoth runtimes25 / 35

Ponytail skill

A minimalism ruleset for coding agents: it pushes the agent toward the smallest correct change and away from speculative abstraction, with the stated philosophy that the best code is the code you never wrote.

What it does for you: Axion builds accumulate scope fast when an agent over engineers a quick fix into a new abstraction layer. Dropping Ponytail's rules into a project's instructions could trim that habit before it costs YY review time, and because it is just text and config, not a running service, the risk of adopting it is low. Worth a side by side trial on one live Axion build to see if output actually gets leaner.

In practice: Reads like a genuine engineering philosophy someone distilled from real frustration, not a gimmick repo. The 162.7k star count climbed fast, which is itself worth treating with a little caution until it settles.

For: Both runtimes. Ships as a plain ruleset and CLI pattern rather than a Claude specific plugin, so it works inside any coding agent harness, Claude Code or Hermes included.

Security4

Quality3

Auditability4

Useful to you3

Useful to community4

Buildable now4

Hermes3

Verdict: test first. Low adoption risk, since it only changes agent behavior rather than running code, but the viral growth is recent enough that its real world effect on build quality has not been proven yet.

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

Source · GitHub About panel confirmed live on 2026-10-13: 162.7k stars, MIT license.

#2

test firstClaude (Axion)24 / 35

blender-kiln plugin

A Claude Code plugin that turns a text brief or a photo into a production ready 3D asset through Blender: sourcing, cleanup, texturing from CC0 scans, optimizing to GLB, FBX or USDZ, rigging and animating, running on either of the two public Blender MCP servers.

What it does for you: This sits directly on top of what YY's own blender-generated-assets-proof skill already assumes exists: a working Blender MCP pipeline that turns a brief into a checked asset. Instead of hand building that pipeline, blender-kiln gives a maintained, versioned plugin to test against the same proof requirements. If the output quality holds up, it could become the actual engine behind that skill rather than a parallel build.

In practice: Clearly built by someone using it for real work, not a demo: the README talks about grading against the real object and batch mode for production runs, and CI badges for both general checks and Blender specific checks are green.

For: Claude (Axion). Ships as a Claude Code skill and plugin with a Claude specific SKILL.md, installed through Claude's own plugin system rather than as a standalone cross runtime tool.

Security3

Quality3

Auditability4

Useful to you5

Useful to community3

Buildable now4

Hermes2

Verdict: test first. Direct overlap with an existing YY skill makes this worth a real trial, but it is not yet installed or proven inside Axion's own workflow, so it should not be called a build until one real asset clears the proof bar.

Install: in Claude Code run /plugin marketplace add elithril/blender-kiln then /plugin install blender-kiln@blender-kiln, not currently in YY's enabledPlugins list per a read only check of ~/.claude/settings.json on 2026-10-14. Review before enabling. Copy

Source · Latest tagged release v2.0.1 with a v2.0.2 entry following soon after, MIT license confirmed on the repo page, both the general and Blender specific GitHub Actions checks show passing.

#3

newtest firstBoth runtimes24 / 35

cua repo

Open source infrastructure for computer use agents: sandboxed virtual machines, a single SDK across macOS, Windows and Linux, a benchmarking suite, and a virtualization layer called Lume, built to scale desktop control across fleets of machines rather than one off demos.

What it does for you: YY has repeatedly needed an agent to click through a real screen rather than call an API: the GHL dashboard is read only and UI only, Sheets work has been flagged as unviable through Chrome MCP for live edits, and several funnel checks require literally clicking through a page. Cua is built for exactly that kind of sandboxed, repeatable desktop control, and because it ships cross platform SDKs it would not lock Axion into one OS.

In practice: Reads like serious infrastructure, not a toy: a Y Combinator backed team, a real benchmarking suite, and a license structure that keeps the core MIT while gating one advanced piece, Cua Spaces, under a source available license.

For: Both runtimes. Ships SDKs in Python, TypeScript, Swift and Kotlin plus a CLI and daemon, so it is callable from any runtime including a Hermes side process, not tied to Claude Code.

Security3

Quality4

Auditability3

Useful to you4

Useful to community4

Buildable now3

Hermes3

Verdict: test first. Directly relevant to several of YY's recurring automation pain points, but sandboxed desktop control carries real blast radius if misconfigured, so it needs a contained trial before touching anything live like GHL or Sheets.

Build #3 cua: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/trycua/cua. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · GitHub About panel confirmed live on 2026-10-13: 29.6k stars, 2.1k forks, MIT core license with Cua Spaces under a separate source available license.

#4

newtest firstBoth runtimes24 / 35

archprint plugin

Scans a TypeScript repository's real import graph, infers the architectural boundaries the code already follows, gates each one on statistical evidence, and emits ready to install ESLint, dependency-cruiser or ts-arch rules so architecture drift gets caught in CI rather than just documented.

What it does for you: Axion's own codebase is mostly Python scripts and skills today, so the direct fit is modest, but any TypeScript heavy client or internal tool, including a future AxionOS or hub build, could use this to catch architecture drift automatically instead of relying on a written rule nobody enforces. The read only remote MCP mode means an agent can check a repo's boundaries before finishing a task without installing anything locally.

In practice: The validation claim, a scan across 92,861 real public TypeScript repositories with zero crashes, is the kind of concrete number that is hard to fake and easy to check, which makes this feel more trustworthy than most new dev tooling.

For: Both runtimes. Runs as a CLI and an optional remote MCP server, both callable from any MCP capable client, so it is not Claude Code exclusive.

Security4

Quality4

Auditability4

Useful to you3

Useful to community3

Buildable now4

Hermes2

Verdict: test first. Low security exposure since the MCP mode is read only and clones to a temp directory it deletes, but there is no current TypeScript project at Axion scale to run it against yet, so it stays a trial rather than a build.

Install: npx archprint scan locally, or run the hosted remote MCP server (archprint mcp --http) and add it to an MCP client, not currently in YY's enabledPlugins list per a read only check of ~/.claude/settings.json on 2026-10-14. Review before enabling. Copy

Source · Tagged releases run from v0.7.0 through v0.11.0 on GitHub, with a companion GitHub Action, Tommkruix/archprint-action, published on the Marketplace.

#5

test firstBoth runtimes23 / 35

Agency Agents plugin

A roster of 298 AI agent personas across 19 divisions, from frontend specialists to community moderators, each with a defined voice, process and deliverables, installable through a small native desktop app that browses the roster and configures the chosen personas into whichever coding agent tool is already in use.

What it does for you: YY already has a claude-subagent-library skill in progress and a stated plan to keep the orchestrator and worker hierarchy at two levels. This roster could shortcut a lot of persona writing by giving a starting library to prune down to the ones that actually fit Axion and MCL work, rather than writing 298 personas from scratch.

In practice: Confident and well packaged, with a proper native app rather than a README full of copy paste blocks, but 298 personas is a lot of surface area and most teams only end up using a handful.

For: Both runtimes. The native installer configures persona files into Claude Code, Cursor, Codex, Gemini CLI, OpenCode, Qwen and Osaurus alike, so it spans runtimes rather than belonging to one.

Security3

Quality3

Auditability3

Useful to you4

Useful to community4

Buildable now4

Hermes2

Verdict: test first. A real shortcut for building out a subagent roster, but adopting all 298 personas unreviewed would violate YY's own rule to keep splits small and verifiable, so this needs a curation pass before anything gets used.

Install: brew install --cask msitarzewski/agency-agents/agency-agents, then run brew trust if Homebrew blocks the untrusted tap, not currently in YY's enabledPlugins list per a read only check of ~/.claude/settings.json on 2026-10-14. Review before enabling. Copy

Source · GitHub About panel confirmed live on 2026-10-13: 159.9k stars, MIT license, installer available via brew install --cask msitarzewski/agency-agents/agency-agents.

#6

test firstBoth runtimes23 / 35

Opik repo

An open source platform for tracing, evaluating and monitoring LLM applications, self hostable under Apache 2.0, covering the same ground as Langfuse and promptfoo but from Comet.

What it does for you: YY already has prompt-eval-promptfoo and llm-observability-openllmetry skills in the stack. Opik would not replace those so much as give a second, Apache 2.0 licensed option to compare eval and tracing output against, which is useful when deciding which tool's judgments to actually trust.

In practice: Established and actively patched rather than flashy, which is the right feel for something that would sit in a production eval pipeline.

For: Both runtimes. Self hosted platform with SDKs across languages, called over an API, so it fits any runtime that can make a network call.

Security4

Quality4

Auditability3

Useful to you3

Useful to community4

Buildable now3

Hermes2

Verdict: test first. Not a new finding, but the patch cadence, three releases inside a week, shows it is being kept current, and a second eval tool is cheap to trial against the ones already in use.

Build #6 Opik: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/comet-ml/opik. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · Recent tagged releases 2.2.39, 2.2.41, 2.2.43 and 2.2.44 in quick succession on GitHub, Apache 2.0 license confirmed on the repo page.

#7

newtest firstClaude (Axion)22 / 35

session-pack skill

A Claude Code skill that moves a session between people or machines with pack and unpack commands, or tracks a conversation in git alongside the code it produced with init, sync, render and why commands.

What it does for you: This is close kin to YY's own handoff skill, which already writes session handoff documents, but session-pack goes further by letting the raw session itself travel between machines or live versioned in git next to the code. Worth comparing side by side with the handoff skill to see whether it should replace, extend, or stay separate from that existing workflow.

In practice: Small, single purpose, and clearly brand new: one star, created a little over a month ago, so this is early days for trusting it with anything important.

For: Claude (Axion). Packs and restores Claude Code's own session file format specifically, so it only makes sense inside that runtime.

Security4

Quality2

Auditability4

Useful to you4

Useful to community2

Buildable now4

Hermes2

Verdict: test first. Directly overlaps with an existing YY workflow, handoffs, low risk to try since it is just packing files, but one star and a few weeks of age means it has not been proven at any real scale yet.

Build #7 session-pack: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/kimbjunn/session-pack. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · GitHub About panel confirmed live on 2026-10-13: 1 star, MIT license, created 2026-09-09, Python.

#8

watchBoth runtimes22 / 35

Agno repo

An SDK for building agent platforms: build agents in any framework, run them as production services, and manage all of them from one control plane. Agno is the successor project to the earlier Phidata framework.

What it does for you: Another multi agent orchestration option alongside the Claude Agent SDK already in use. The main reason to look twice is the jump to a v3 major version, which usually means a real architecture change worth understanding even if Axion does not switch to it.

In practice: Mature or active rather than new: 49 repositories under the org and a steady release cadence, this is not a weekend project.

For: Both runtimes. Python SDK that is model agnostic and can run inside any process, Axion script or a Hermes side service alike.

Security3

Quality4

Auditability3

Useful to you3

Useful to community4

Buildable now3

Hermes2

Verdict: watch. A major version bump is worth knowing about, but Axion is not short an orchestration framework right now, so this stays a watch item rather than something to trial immediately.

Build #8 Agno: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/agno-agi/agno. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · GitHub release history shows a jump from the v2.7.x series to tag 3.0.1, a major version change, confirmed via the repo's releases page.

#9

watchBoth runtimes22 / 35

Letta repo

An open source framework and hosted service for stateful agents, where agents read, write and modify their own persistent memory blocks using tools, making memory management part of the agent's own execution loop rather than a bolt on afterward.

What it does for you: Axion's own second brain and context-mode stack already handle a version of this problem. Letta is worth knowing as a comparison point: it treats memory as something the agent actively manages during a task, which is a different model from the vault plus KB index approach Axion currently uses.

In practice: Established framework territory, with its own hosted option alongside the open source core, which usually means real production use behind it.

For: Both runtimes. A self hosted stateful agent server with a REST API, callable from any language or runtime, Hermes included.

Security3

Quality4

Auditability3

Useful to you3

Useful to community4

Buildable now3

Hermes2

Verdict: watch. A useful comparison point for how Axion's own memory architecture could evolve, but there is no concrete gap it fills today that the brain vaults and context-mode do not already cover.

Build #9 Letta: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/letta-ai/letta. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · GitHub release history confirms a recent patch tag, 0.16.8, alongside a separate letta-code repository under the same org.

#10

newwatchClaude (Axion)21 / 35

cc-pr-tracker plugin

A Claude Code mod that watches a GitHub pull request from inside a session: merge state, review decision and required checks shown as one line above the prompt, refreshed every minute, with a toast and sound when something changes and a note passed to Claude when a check fails.

What it does for you: The Skills repo is the one piece of Axion that is actually a tracked git repo with a remote. If that repo starts taking pull requests, a mod like this would let an agent session notice a failing check or a merge without YY having to tab over to GitHub to look.

In practice: Small and genuinely useful rather than flashy, the kind of focused tool that earns its keep quietly instead of chasing stars.

For: Claude (Axion). A Claude Code mod: its hooks run inside Claude Code's own process specifically and draw into its interface, so it only works in that runtime.

Security4

Quality2

Auditability4

Useful to you3

Useful to community3

Buildable now4

Hermes1

Verdict: watch. A real fit for a repo with active PR review, but the Skills repo does not currently run that workflow, so this waits until there is a PR flow to watch.

Install: inside Claude Code run /plugin to add sezaakgun/cc-pr-tracker, not currently in YY's enabledPlugins list per a read only check of ~/.claude/settings.json on 2026-10-14. Review before enabling. Copy

Source · GitHub repo data confirmed: 5 stars and 2 forks, built as a Claude Code mod using TypeScript hooks rather than shell command hooks.

#11

watchStandalone tool21 / 35

Activepieces repo

An open source n8n style workflow automation platform whose building blocks, called pieces, support the Model Context Protocol directly, meaning a workflow built in it can be consumed by an LLM client as a set of callable tools.

What it does for you: If YY ever wants a visual workflow builder for something less code heavy than the Telegram bridge's own Python routines, MCP support means that workflow could be exposed to Claude or Hermes as tools rather than living in total isolation.

In practice: Solidly built open source infrastructure rather than a wrapper around someone else's API, with real MCP depth rather than a bolt on integration.

For: Standalone tool. A hosted or self hosted workflow builder YY would run and configure directly, not a library to embed inside Axion's own code.

Security3

Quality4

Auditability2

Useful to you3

Useful to community4

Buildable now3

Hermes2

Verdict: watch. A reasonable n8n alternative with a real MCP angle, but the Telegram bridge already covers YY's current automation needs, so there is no active gap for this to fill yet.

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

Source · GitHub About panel confirmed live on 2026-10-14: 25.0k stars, 4.3k forks, with the repo's own description citing roughly 400 MCP servers available for its AI agents and workflows.

#12

newwatchStandalone tool21 / 35

pi-mono repo

An AI agent toolkit built around a minimal, token efficient architecture: a coding agent CLI, a unified LLM API across providers, and TUI and web UI libraries, created by Mario Zechner and now stewarded by Earendil Inc, the public benefit corporation founded by Flask creator Armin Ronacher.

What it does for you: A genuine alternative to the Claude Code harness itself, built specifically to minimize token use per action. Worth a side by side comparison on one real task to see whether the token efficiency claim translates into a meaningfully cheaper or faster build for Axion work, before considering any switch.

In practice: Built by people with real credibility in developer tooling, and the minimalist design philosophy comes through as deliberate rather than a feature gap.

For: Standalone tool. A separate coding agent harness and CLI that YY would run instead of Claude Code, not something embedded inside it.

Security3

Quality4

Auditability3

Useful to you2

Useful to community4

Buildable now3

Hermes2

Verdict: watch. Credible team and a real efficiency angle, but switching core harnesses is a heavy decision that should not follow from one scouting pass; this stays a watch item pending a dedicated comparison.

Build #12 pi-mono: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/badlogic/pi-mono. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · GitHub About panel confirmed live on 2026-10-13: 114.9k stars, 14.6k forks, latest tagged version 1.0.4 dated 2026-10-05.

#13

newwatchBoth runtimes20 / 35

Dia2 repo

An open weights, streaming dialogue text to speech model from Nari Labs, described in community posts as Dia2, built to generate speech without needing a full sentence first, which suits low latency speech to speech systems. The original Dia model is a 1.6B parameter two speaker dialogue TTS released under Apache 2.0.

What it does for you: Voice cloning and realistic dialogue synthesis touches several things on YY's own list: Marc's AI clone performance work, ai-talking-video-editing, and the speak-marc and speak-yy voice skills. A lower latency, streaming TTS model could matter for any of those if the quality holds up, but this needs a listening test before it goes near anything client facing given the existing rule that Marc's voice and likeness consent is not fully recorded yet.

In practice: The underlying Dia work is credible and widely cited, but the specific Dia2 branding and repository are murkier than the rest of this issue, worth treating the sourcing gap honestly rather than papering over it.

For: Both runtimes. Model weights callable from any local Python inference stack, not tied to one agent runtime.

Security4

Quality3

Auditability3

Useful to you3

Useful to community4

Buildable now2

Hermes1

Verdict: watch. Real potential fit for voice work, but the exact repository for the Dia2 branding could not be pinned down with confidence this run, so it stays a watch item until that is resolved.

Build #13 Dia2: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/nari-labs/dia. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · The base Dia project, nari-labs/dia, is confirmed on GitHub under Apache 2.0; the Dia2 streaming variant is referenced in a Hacker News Show HN post rather than a dedicated GitHub release this run.

#14

newwatchStandalone tool19 / 35

Unmute repo

A compiler for voice agents: write a package of YAML and Markdown describing an agent's identity, models, tools and workflow, and Unmute turns it into a real Python project for Pipecat or LiveKit, with pinned dependencies, a Dockerfile and a runbook, and no runtime dependency on Unmute itself once compiled.

What it does for you: If YY ever needs a production voice agent for MCL, for example automated webinar reminder calls or a phone based intake flow, this gives a path from a simple spec to code that is actually owned and deployable, rather than staying locked to one vendor's hosted voice platform.

In practice: A sensible, infrastructure minded approach: compile once to code you own rather than depend on a service forever, which fits how YY generally prefers tools that do not create ongoing vendor lock in.

For: Standalone tool. Compiles to a separate Pipecat or LiveKit codebase that runs outside Claude Code or Hermes entirely.

Security3

Quality3

Auditability3

Useful to you3

Useful to community3

Buildable now2

Hermes2

Verdict: watch. No current MCL workflow needs a production voice agent yet, so this is worth knowing about rather than trialing immediately.

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

Source · GitHub confirms an active release history through tag v0.7.0, with recent pull requests adding realtime architecture support.

#15

newwatchStandalone tool19 / 35

AnythingLLM repo

A self hostable, local first application that bundles chat, RAG document search, and agent skills including web browsing and image generation into one on device tool, with v1.17.0 released 2026-10-01 adding a new local LLM provider option and web search with zero data retention.

What it does for you: Axion already has a context-mode and second brain stack covering local RAG and knowledge retrieval, so the direct gap this fills is small. Still worth knowing as a packaged alternative if YY ever wants a simpler, single app local RAG tool for a narrower one off project rather than the full brain vault setup.

In practice: Polished and feature dense, clearly aimed at end users rather than developers building their own stack, which is exactly why it overlaps with rather than extends Axion's existing architecture.

For: Standalone tool. A separate self hosted application with its own UI that YY would run alongside Axion, not code embedded into it.

Security3

Quality4

Auditability2

Useful to you2

Useful to community4

Buildable now3

Hermes1

Verdict: watch. Actively maintained with a real October release, but it duplicates rather than extends what Axion's context-mode and brain vaults already do, so it stays a watch item.

Build #15 AnythingLLM: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/Mintplex-Labs/anything-llm. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · Release v1.17.0 confirmed dated 2026-10-01, adding a new local LLM provider option, Google Vertex AI support, and a generate-image agent skill.

#16

watchStandalone tool18 / 35

Hermes Agent repo

An open source, self improving agent framework from Nous Research under the MIT license, shipping roughly 160 built in skills covering GitHub workflows, document handling and research, reaching users across 20 messaging platforms, with an enterprise tier called Hermes for Businesses announced alongside a Series B funding round on 2026-10-07.

What it does for you: This is worth flagging for one reason above all others: it shares the name Hermes with YY's own Axion Telegram bridge. They are completely unrelated projects, but any future mention of Hermes needs a split second check on which one is meant, in conversation and in any written note.

In practice: A real, well funded project with genuine scale, not a side project, which makes the name collision more likely to cause confusion over time rather than less, since it will keep showing up in searches.

For: Standalone tool. A separate self hosted agent product with its own skill library and messaging integrations, confusingly similar in name to YY's own Hermes bridge but otherwise unrelated.

Security3

Quality4

Auditability2

Useful to you2

Useful to community4

Buildable now2

Hermes1

Verdict: watch. Technically capable and well resourced, but the name collision with YY's own Hermes bridge is a real practical hazard that should be flagged every time this comes up rather than adopted casually.

Build #16 Hermes Agent: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/NousResearch/hermes-agent. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · GitHub About panel confirmed live on 2026-10-13: 253.5k stars, MIT license, with a tagged release v2026.7.20 visible in release history.

#17

newwatchBoth runtimes18 / 35

REA plugin

An MCP server and CLI, distributed as the npm package rea-agents, that gives coding agents tools to reverse engineer software, from app behavior down to native binaries, without needing access to the original source code.

What it does for you: Nothing in YY's current MCL or Axion work calls for binary reverse engineering, so the direct fit is low despite the scale of this project's growth. It is included here mainly because the growth itself is the story: from relatively unnoticed in April 2026 to over 200,000 stars by mid October, one of the fastest climbs GitHub has seen.

In practice: The project's own license file is unusually careful about the line between its own MIT code and the commercial decompilers it can orchestrate, IDA Pro, Hopper, JEB, explicitly stating it grants no license to those tools, which reads as a team aware of the legal edge they are operating near.

For: Both runtimes. An MCP server and CLI usable by any MCP capable client, Claude Code or Hermes included, not restricted to one runtime.

Security2

Quality4

Auditability3

Useful to you1

Useful to community5

Buildable now2

Hermes1

Verdict: watch. A genuinely notable growth story worth knowing about, but low direct usefulness to YY's business and real legal exposure around commercial decompiler licensing keep this at watch rather than anything closer to adoption.

Install: npx rea-agents for the CLI, or add the REA MCP server to an MCP capable client, not currently in YY's enabledPlugins list per a read only check of ~/.claude/settings.json on 2026-10-14. Review before enabling, and confirm licensing terms for any commercial decompiler used alongside it. Copy

Source · GitHub About panel confirmed live on 2026-10-13: 204.5k stars, up from a 180,000 star milestone noted in the project's own star history section days earlier, MIT license with a separate third party notices file for bundled tooling.

#18

newwatchClaude (Axion)17 / 35

pwa2play plugin

A Claude Code plugin that packages a progressive web app into a signed, Play Store ready Android app bundle using Google's open source Bubblewrap tool, checking target API level, signing key reuse, assetlinks fingerprints and version codes before upload.

What it does for you: Only relevant if a future project, for example a mobile facing piece of AxionOS or a client funnel, needs an actual Play Store listing from a web app rather than a bookmarked page. Nothing on YY's current project list calls for this yet, so it is a tool to remember rather than one to try this week.

In practice: Narrow and clearly purpose built, the kind of plugin that solves one specific, annoying packaging problem rather than trying to be a platform.

For: Claude (Axion). Ships as a Claude Code plugin invoked through Claude's own plugin system.

Security3

Quality2

Auditability3

Useful to you2

Useful to community3

Buildable now3

Hermes1

Verdict: watch. No current YY project needs a Play Store bundle, so this is filed for later rather than trialed now.

Install: in Claude Code run /plugin marketplace add degordonstech/pwa2play then /plugin install pwa2play, not currently in YY's enabledPlugins list per a read only check of ~/.claude/settings.json on 2026-10-14. Review before enabling. Copy

Source · Added to the awesome-claude-code directory on 2026-10-07, built on Google's open source Bubblewrap toolchain.

#19

newskipBoth runtimes16 / 35

Hostinger MCP Server plugin

An official, vendor hosted remote MCP server from Hostinger that exposes its own API, Websites, VPS, Domains, Email Marketing, Subscriptions and Payments, and Ecommerce, as callable tools to any MCP compatible client, with no local installation required.

What it does for you: Nothing in YY's current stack runs on Hostinger: hosting, deployment and DNS for YY's own sites run through Cloudflare Pages and other providers per the AXION inventory. Included here mainly as a pattern worth noting, an official vendor shipping its own remote MCP server directly rather than leaving it to the community, which more hosting and SaaS vendors are likely to copy.

In practice: Clean, official, and low effort to connect, exactly what a vendor hosted MCP server should look like, but there is no account behind it on YY's side to make it useful today.

For: Both runtimes. A remote MCP server callable by any MCP client, Claude Code or a Hermes side agent alike, with no local install needed.

Security3

Quality3

Auditability2

Useful to you1

Useful to community3

Buildable now3

Hermes1

Verdict: skip. No matching Hostinger account exists anywhere in YY's current stack, so there is nothing to connect this to; noted for the pattern, not for actual use.

Install: add the remote MCP endpoint https://mcp.hostinger.com to an MCP capable client, no local install or API key needed, not currently in YY's enabledPlugins list per a read only check of ~/.claude/settings.json on 2026-10-14. Review before enabling, and note no matching Hostinger account exists in YY's current stack. Copy

Source · Official documentation confirms the server runs at https://mcp.hostinger.com over Streamable HTTP, exposing 6 API categories, Websites, VPS, Domains, Email Marketing, Subscriptions and Payments, and Ecommerce, with no API key or account setup step required to connect, per Hostinger's own support page.

#20

newskipStandalone tool15 / 35

cmux repo

An open source, Ghostty based macOS terminal with vertical tabs and notifications purpose built for running and tracking several parallel AI coding agent sessions at once.

What it does for you: A genuinely interesting idea for anyone juggling parallel agent sessions, but YY's fleet runs on Windows 11 with PowerShell 5.1 as the primary shell, and this is macOS only with Linux and Windows ports still described as in progress. There is no way to use this today on YY's own machine.

In practice: A thoughtfully designed tool for a real problem, multitasking across agent sessions, let down for YY's purposes only by platform timing rather than by the idea itself.

For: Standalone tool. A separate macOS terminal application YY would install and run, not a library or plugin embedded into Axion or Hermes.

Security3

Quality3

Auditability3

Useful to you1

Useful to community3

Buildable now1

Hermes1

Verdict: skip. macOS only today against a Windows primary machine is a hard platform mismatch, not a quality problem; revisit once Windows support actually ships rather than now.

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

Source · GitHub About panel confirmed live on 2026-10-13: 28.2k stars, 2.5k forks, built in Swift, with the project's own notes describing Linux and Windows ports as still in progress.