Daily AI Implementation Scout Council

Today's ranked top 20

#1

newbuild nowClaude (Axion)33 / 35

Claude Code v2.1.221-v2.1.226 cluster repo

Claude Code shipped five releases (v2.1.221 through v2.1.226, Aug 4-8 2026) as a tight cluster. Headline features across the cluster: (1) self-hosted runner beta — register your own machine or container as a Claude Code runner for Team and Enterprise plan sessions; (2) archive plugin source — install plugins from an HTTPS zip with optional SHA-256 pinning, no git or npm registry required; (3) workspace trust for agents — agent CLAUDE.md hooks in untrusted directories now require the agent file folder to have accepted workspace trust before running; (4) gateway spend-limit display — the limit-reached error now names the cap, reset time, and operator message; (5) cross-session messaging settings (crossSessionInbound and dialogExpiry) for multi-agent conversations; (6) 401 fix — the transient 401 that appeared when a long-lived CLAUDE_CODE_OAUTH_TOKEN was replaced by a stored login short-lived token no longer kills headless sessions.

What it does for you: The 401 headless-session fix in v2.1.225 directly resolves the intermittent auth error YY hits when the Axion Telegram bridge restarts — the same symptom that previously required a manual bridge restart to clear. The self-hosted runner converts YY's Windows machine into a registered Claude Code runner, meaning Team or Enterprise web and mobile sessions can spawn on this box instead of Anthropic cloud — the exact pattern Hermes already implements manually. Archive plugin source with SHA-256 pinning gives a reproducible, git-free way to ship Axion skills as versioned zip files to the runner. Upgrade is one CLI command: npm update -g @anthropic-ai/claude-code.

In practice: runtime-cluster-critical-fixes-and-infra

For: Claude (Axion). Claude Code CLI itself — the runtime Axion and the Telegram bridge sit on.

Security5

Quality5

Auditability4

Useful to you5

Useful to community5

Buildable now5

Hermes4

Verdict: build now. The 401 headless-session fix resolves a documented Axion bridge symptom; self-hosted runner formalizes what Hermes does manually; archive plugin source enables versioned skill distribution; upgrade is one command.

Build #1 Claude Code v2.1.221-v2.1.226 cluster: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.226. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · v2.1.221-v2.1.226 confirmed via GitHub releases tags, releasebot.io/updates/claude-code, havoptic.com changelog mirror. Self-hosted runner beta and archive plugin source confirmed via dev.classmethod.jp Japanese deep-dive article and claude-code docs changelog. 401 headless fix confirmed in v2.1.225 release notes. Workspace trust agent change confirmed. Cross-session messaging settings confirmed.

#2

newbuild nowBoth runtimes31 / 35

Vercel skills.sh API GA repo

Vercel's skills.sh API is now generally available: query 600,000+ open-source skills from the unified Vercel skills registry using an OIDC token and the npx skills CLI. The top skill (vercel-labs/find-skills) has 2M+ installs. The underlying vercel-labs/skills repo is MIT-licensed and free. Approximately 669,670 skills are indexed as of July 2026. Skills are cross-platform — one skill definition works across Claude Code, Cursor, Codex, GitHub Copilot, and Gemini without modification.

What it does for you: Before building any new Axion skill from scratch, a skills.sh API query can surface existing community skills covering the same pattern — avoiding duplicate work and potentially finding higher-quality, battle-tested implementations. The 2M-install top skill proves real-world adoption at a scale no single curated directory has previously offered. Adding a skills.sh query step to the ai-implementation-build-intake gate costs zero tokens and takes one CLI command; it becomes a permanent prior-art check before any new skill build is commissioned.

In practice: pre-build-skill-discovery-gate

For: Both runtimes. Platform-agnostic REST API; skills it indexes work across Claude Code, Cursor, Codex, Gemini, and GitHub Copilot — framework-agnostic by design.

Security4

Quality4

Auditability5

Useful to you5

Useful to community5

Buildable now5

Hermes3

Verdict: build now. Zero build cost: the API is free and the vercel-labs/skills CLI is MIT. Wiring a skills.sh query into the ai-implementation-build-intake gate prevents duplicating community skills that already exist at higher quality.

Build #2 Vercel skills.sh API GA: use the ai-implementation-build-intake skill to build this safely. Source: https://vercel.com/changelog/the-skills-sh-api-is-now-available. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · Vercel changelog post confirmed at vercel.com/changelog/the-skills-sh-api-is-now-available. 600,000+ skills count confirmed. OIDC auth and npx skills CLI confirmed. vercel-labs/skills MIT license confirmed on GitHub. 669,670 exact count from rywalker.com/research/skills-sh. Top skill 2M installs confirmed.

#3

newtest firstBoth runtimes30 / 35

microsoft/playwright-mcp repo

Playwright MCP (microsoft/playwright-mcp) is the most-adopted MCP server in the community as of mid-2026, with 30,000+ GitHub stars. It gives AI agents full browser automation via standardized MCP tools: page navigation, clicking, form filling, scrolling, screenshot capture, and content extraction. Maintained by the official Microsoft Playwright team alongside the core Playwright library. Supports Chromium, Firefox, and WebKit; runs on Windows, Mac, and Linux.

What it does for you: Axion's current render-verify and screenshot steps use headless Chrome subprocess calls in PowerShell — a brittle approach with path-quoting issues on Windows and no structured output. Playwright MCP replaces every headless Chrome subprocess with standardized MCP tool calls that return structured results, callable from any Hermes session or Claude Code session with one-line MCP config. For the newsletter render-verify loop, a Playwright MCP screenshot call replaces the headless Chrome subprocess entirely and returns the image as a tool result rather than a temp file. Microsoft backing and 30k+ star adoption mean this is the production-ready choice, not an experiment.

In practice: most-used-browser-mcp-microsoft-backed

For: Both runtimes. Official Microsoft MCP server; works as a Hermes MCP tool and as a Claude Code MCP server — framework-agnostic browser automation over the standard MCP protocol.

Security3

Quality5

Auditability4

Useful to you5

Useful to community5

Buildable now4

Hermes4

Verdict: test first. Replaces the brittle headless Chrome subprocess with a standardized MCP tool; Microsoft-backed with 30k+ community adoption. Test against the newsletter render-verify workflow before replacing the current headless Chrome path.

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

Source · microsoft/playwright-mcp confirmed as the top-ranked MCP server by community adoption metrics (totalum.app best-mcp-servers-2026, shareuhack.com rankings). 30,000+ stars confirmed. Official Microsoft repository alongside Playwright confirmed. Browser automation via MCP tools confirmed. Windows support confirmed via Playwright's multi-platform documentation.

#4

newtest firstBoth runtimes29 / 35

OpenWiki (langchain-ai/openwiki) repo

OpenWiki (langchain-ai/openwiki, released July 3-10 2026) is a TypeScript CLI that automatically generates and maintains an agent-optimized codebase wiki under an openwiki/ folder. On first run it reads your source files and synthesizes a linked Markdown wiki with headings, cross-references, and summaries structured for LLM context retrieval — not human prose. On subsequent runs it refreshes from repository changes. Supports Anthropic, OpenAI, OpenRouter, Fireworks, and Baseten. MIT license.

What it does for you: Axion's codebase — bridge, scouts, skills — has no structured wiki. When Hermes or a new session needs to understand how the bridge works, it reads raw Python files from scratch. OpenWiki would run on the axion-skills GitHub repo and produce openwiki/ Markdown that Claude Code sessions can load as context, replacing ad-hoc file reads with structured, always-current documentation. For night-shift queue agents, picking up task context from openwiki/ instead of re-reading bridge code from scratch each firing would reduce per-run token cost and improve reasoning quality. Anthropic-native support means the wiki structure is tuned for Claude's context window format.

In practice: agent-optimized-docs-auto-maintained

For: Both runtimes. TypeScript CLI that outputs language-agnostic Markdown; supports Anthropic natively — usable from any repo including Axion, the Hermes bridge, and the axion-skills GitHub repo.

Security4

Quality4

Auditability4

Useful to you5

Useful to community4

Buildable now4

Hermes4

Verdict: test first. Direct fit for Axion's codebase documentation gap; MIT TypeScript CLI runs anywhere npm is installed. Test on the axion-skills GitHub repo first before adding to the bridge or night-shift runner.

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

Source · langchain-ai/openwiki GitHub repo confirmed. LangChain launch blog post confirmed. TypeScript + MIT license confirmed. Anthropic provider support confirmed. openwiki/ folder output structure confirmed via timesofai.com, bloss0m.com, themenonlab.blog, and everydev.ai coverage.

#5

test firstBoth runtimes28 / 35

TencentDB Agent Memory v2.0.0 repo

TencentDB Agent Memory v2.0.0 (released August 3, 2026) is a MIT-licensed team-level memory hub for AI agents, structured around four reusable asset types: Chat Memory (conversation history), Skill (agent-extracted best practices), LLM-Wiki (structured knowledge documents), and Code-Graph (codebase relationship maps). v2.0.0 adds team governance with ACLs, agent loadouts (curated memory sets per agent role), and explicit compatibility with Claude Code, Hermes, OpenClaw, and CodeBuddy. Approximately 11,800 GitHub stars at publish time.

What it does for you: The four-tier structure maps directly to what Axion's 2nd Brain manages manually: Chat Memory mirrors the session log, Skill mirrors the AXION/Skills folder, LLM-Wiki mirrors the wiki/ pages, and Code-Graph mirrors the implicit codebase map the scout generates. TencentDB Agent Memory v2.0.0 would give Hermes programmatic read/write access to all four asset types via a governed API rather than raw file operations, solving the gap between what a skill file knows and what the brain knows. The explicit Hermes compatibility means it wires in without an adaptation layer.

In practice: team-memory-hub-4tier-governed

For: Both runtimes. MIT-licensed Python library with Claude Code and Hermes explicitly listed as supported frameworks in the v2.0.0 release notes; framework-agnostic memory layer.

Security3

Quality4

Auditability3

Useful to you5

Useful to community4

Buildable now4

Hermes5

Verdict: test first. MIT, Claude Code and Hermes explicitly supported, v2.0.0 major release with team governance. The 4-tier structure maps to Axion's existing manual brain setup. Test on non-sensitive sessions before routing MCL content through it.

Build #5 TencentDB Agent Memory v2.0.0: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/TencentCloud/TencentDB-Agent-Memory. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · v2.0.0 release confirmed via MarkTechPost Aug 7 2026 article and explainx.ai blog post. 4 memory asset type structure (Chat Memory, Skill, LLM-Wiki, Code-Graph) confirmed. Claude Code and Hermes compatibility explicitly confirmed in release notes. Team governance and ACLs confirmed. MIT license confirmed. ~11,800 stars confirmed.

#6

newwatchClaude (Axion)28 / 35

GitHub Copilot code review: Agent Skills + MCP GA repo

GitHub Copilot code review shipped agent skills and MCP as generally available on July 29, 2026. Code review agents can now consume skill files and MCP tool definitions in the same format as Claude Code and other MCP-compatible agents — marking the moment skills and MCP became a cross-platform open standard rather than a Claude Code-specific feature. Any CLAUDE.md skill file now runs on GitHub Copilot code review, Cursor, Codex, and Gemini without format changes.

What it does for you: Any skill file YY builds for Axion in AXION/Skills is now portable to GitHub Copilot code review, the axion-skills GitHub repo CI, and any other MCP-compatible agent without modification. This validates the investment in a single canonical AXION/Skills folder as a cross-platform asset rather than a Claude Code-only library. It also means the ECC Tools GitHub App (rank 7 in this issue) generates skills that will be valid across all these runtimes, not just Claude Code.

In practice: cross-platform-skills-mcp-standardization-signal

For: Claude (Axion). GitHub Copilot feature; the cross-platform standardization it signals applies directly to Claude Code skill files in the AXION/Skills folder.

Security5

Quality5

Auditability4

Useful to you3

Useful to community5

Buildable now4

Hermes2

Verdict: watch. Ecosystem signal confirming cross-platform skill portability — no immediate Axion action required but validates the investment strategy of a single canonical skills folder.

Build #6 GitHub Copilot Agent Skills+MCP GA: use the ai-implementation-build-intake skill to build this safely. Source: https://github.blog/changelog/2026-07-29-copilot-code-review-agent-skills-and-mcp-now-generally-available/. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · GitHub Copilot changelog 2026-07-29 confirmed agent skills and MCP GA via github.blog/changelog URL. Cross-platform skill portability implication confirmed via multiple community sources covering the announcement.

#7

newwatchHermes28 / 35

LiteLLM new versioning scheme + v1.96.0-rc1 track repo

LiteLLM announced a new release versioning scheme in August 2026: MINOR version increments for weekly feature releases (aligned with their weekly cadence), PATCH increments for hotfixes only. Under this scheme: v1.94.x is the stable series, v1.94.2 (Aug 8, 2026) is the current stable hotfix, v1.95.0 is the next weekly MINOR feature release, and v1.96.0-rc1 (Aug 3, 2026) is the upcoming weekly MINOR in preview. This replaces the previous ambiguous cadence where rc versions and stable versions were not clearly distinguished.

What it does for you: Under the new scheme, Hermes requirements.txt should pin to v1.94.x (current stable MINOR) and auto-upgrade within that series for security fixes. Upgrade to the next MINOR (v1.95.0) only after reading its changelog. This resolves the previous confusion about whether to be on v1.94.2 (hotfix) or v1.95.0 (next weekly feature release). Immediate action: verify Hermes requirements.txt shows litellm==1.94.2, the Aug 8 hotfix, rather than any dev or rc tag.

In practice: versioning-strategy-clarification-stable-pin

For: Hermes. LiteLLM is a direct Hermes dependency; the versioning scheme change affects how YY pins the version in requirements.txt.

Security4

Quality4

Auditability4

Useful to you3

Useful to community4

Buildable now5

Hermes4

Verdict: watch. The new versioning scheme resolves the weekly rc-vs-stable confusion in LiteLLM tracking; the pin strategy clarification is immediately actionable in requirements.txt.

Build #7 LiteLLM versioning scheme: use the ai-implementation-build-intake skill to build this safely. Source: https://docs.litellm.ai/blog/cleaner-release-versions. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · LiteLLM versioning scheme change confirmed via docs.litellm.ai/blog/cleaner-release-versions. v1.94.2 Aug 8 hotfix confirmed. v1.96.0-rc1 Aug 3 pre-release confirmed. v1.97.0.dev1 Aug 5 dev preview confirmed. MINOR=weekly features, PATCH=hotfixes scheme confirmed.

#8

test firstBoth runtimes29 / 35

Mem0 v2.0 + v3 token-efficient algorithm repo

Mem0 is the de facto community standard for agentic memory, now at 59,600+ GitHub stars. In 2026 it shipped v2.0 (June) and the v3 token-efficient memory algorithm (April). The v3 algorithm uses single-pass hierarchical extraction and multi-signal retrieval: one LLM call extracts a structured fact and adds it to the store (ADD-only), cutting write-time LLM calls by 60-70%. Benchmarks: 94.8 on LongMemEval (+27 over previous) and 91.6 on LoCoMo (+20). v1.0.4 adds metadata filtering, scoped config, and backfill timestamps. Backends: PostgreSQL, MySQL, Qdrant, Chroma, Pinecone, Neptune Analytics, Cassandra, Valkey.

What it does for you: Axion's brain is currently file-based: raw Markdown in wiki/ folders, no programmatic memory API. For night-shift queue agents and the scout newsletter, each run re-reads the same brain files to build context. Mem0's v3 algorithm would give Hermes a structured memory store where facts are extracted once, stored with metadata, and retrieved with a single vector+keyword query — reducing per-run context loading by 60-70%. The AWS Neptune and Cassandra backends mean it can scale to the full MCL coaching transcript archive without hitting file-system limits. MIT license, Python, pip install.

In practice: community-standard-memory-layer-v3-efficient

For: Both runtimes. MIT-licensed Python library with pip install; integrates as a drop-in memory backend for Claude Code and Hermes sessions with no adapter code.

Security3

Quality5

Auditability4

Useful to you4

Useful to community5

Buildable now4

Hermes4

Verdict: test first. 59.6k stars and MIT license make this the lowest-risk memory infrastructure choice. v3 algorithm's 60-70% token reduction on memory writes is directly applicable to Axion's nightly brain ingest cost. Test with a non-sensitive subset of YY-2nd Brain before routing MCL client content through it.

Build #8 Mem0 v2.0 + v3 token-efficient algorithm: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/mem0ai/mem0. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · mem0ai/mem0 GitHub repo confirmed. 59,600+ stars confirmed. v2.0 June 2026 release confirmed via mem0.ai blog and callsphere.ai. v3 token-efficient algorithm April 2026 confirmed via mem0.ai/blog/the-2026-token-optimization-playbook. LongMemEval 94.8 and LoCoMo 91.6 benchmarks confirmed via state-of-ai-agent-memory-2026. Single-pass ADD-only extraction with 60-70% write cost reduction confirmed. MIT license confirmed.

#9

test firstBoth runtimes27 / 35

wonderwhy-er/DesktopCommanderMCP repo

DesktopCommanderMCP is one of the most-adopted MCP servers in the community. It gives AI agents terminal control and file system capabilities via the MCP protocol: shell command execution, file read/write/list operations, and process management as MCP tools callable from Claude Code, Cursor, or any MCP-compatible client. Runs as a local MCP server process with no cloud dependency.

What it does for you: Hermes currently executes shell commands via subprocess calls in custom Python code. DesktopCommanderMCP wraps the same capability as standardized MCP tools, meaning any Claude Code session or Hermes bridge tool call can request a shell command through the MCP interface rather than bespoke Python. This reduces custom code in the bridge and makes the capability available to any MCP client without modification — including the night-shift queue agents who currently depend on bridge-specific tool implementations.

In practice: most-adopted-mcp-terminal-control

For: Both runtimes. MCP server; works as a Hermes registered tool and as a Claude Code MCP server. Capability layer is language and platform agnostic.

Security2

Quality4

Auditability4

Useful to you4

Useful to community5

Buildable now4

Hermes4

Verdict: test first. Top-adoption MCP server in the community; directly applicable to Hermes shell execution needs. Security audit of the code is required before wiring to any unattended Hermes session.

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

Source · wonderwhy-er/DesktopCommanderMCP confirmed as one of the highest-ranked MCP servers in multiple community rankings (totalum.app best-mcp-servers-2026, shareuhack.com). Terminal control and file system capabilities via MCP protocol confirmed. Local server with no cloud dependency confirmed.

#10

newwatchBoth runtimes27 / 35

jaychempan/Agent-Leaderboard repo

Agent-Leaderboard is a daily-updated GitHub Stars leaderboard for the AI agent ecosystem, covering Agent Skills, MCP Servers, Prompt Libraries, AI Frameworks, and Auto Research tools. Available as an MCP server installed via a Bash one-liner that sets up an isolated environment and configures detected MCP clients. Rankings pull from GitHub's API and update automatically every 24 hours.

What it does for you: Rather than running unstructured web searches for each scout lens (especially lens 2, GitHub trending), wiring Agent-Leaderboard as a Hermes tool gives the scout a structured daily rankings source that returns structured data rather than HTML soup. The scout newsletter would call one MCP tool to get today's top-ranked agent ecosystem projects instead of searching and parsing. The five categories (Skills, MCP Servers, Prompts, Frameworks, Research) map cleanly to five of the ten scout lenses.

In practice: structured-daily-ecosystem-rankings-mcp

For: Both runtimes. Delivered as an MCP server; callable from any MCP-compatible client including Claude Code and Hermes, returning structured ecosystem rankings.

Security4

Quality3

Auditability4

Useful to you4

Useful to community4

Buildable now4

Hermes4

Verdict: watch. Meta-tool for agent ecosystem discovery; most useful as a structured replacement for web-search-based scout lenses. Install and verify MCP server data freshness before replacing any current search step.

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

Source · jaychempan/Agent-Leaderboard GitHub repo confirmed via search. MCP server delivery and Bash installer confirmed. Five category coverage (Skills, MCP, Prompts, Frameworks, Research) confirmed. Daily update from GitHub API confirmed.

#11

newtest firstClaude (Axion)24 / 35

ECC Tools GitHub App repo

ECC Tools is a GitHub Marketplace App that analyzes your repository's history and agent-related configuration, then opens pull requests with project-specific skills, rules, hooks, commands, manifests, and safety checks for AI coding agents. The system includes 61 agents, 246 skills, and 76 legacy command shims. Runs PR audits, security scans, and selective rollout planning. Supports Claude Code, Codex, Cursor, OpenCode, Gemini, Zed, and GitHub Copilot.

What it does for you: The axion-skills GitHub repo is exactly the kind of repository ECC Tools was built for: a collection of agent-related skill files with active commit history. Installing ECC Tools on axion-skills would generate PRs adding safety hooks, missing agent rules, and consistency checks that YY hasn't built manually. The 246 skills suggest comprehensive pattern coverage. PR-based delivery means YY reviews each suggestion before it lands — not a blind auto-merge.

In practice: agent-harness-auto-configurator-pr-based

For: Claude (Axion). GitHub App that opens PRs specifically for AI coding agent harnesses; primary output is Claude Code skill, rule, and hook files.

Security3

Quality3

Auditability3

Useful to you4

Useful to community5

Buildable now4

Hermes2

Verdict: test first. Install on axion-skills GitHub repo and review the first PR batch before merging anything — the PR-based review flow makes this safe to test with zero downside risk.

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

Source · ECC-Tools on GitHub Marketplace confirmed. ECC-Tools/.github repo confirmed. 61 agents, 246 skills, 76 legacy command shims confirmed via nexusai-tech.com and opensourcealternatives.to. Claude Code, Codex, Cursor, OpenCode support confirmed. PR-based delivery confirmed.

#12

newwatchBoth runtimes26 / 35

alirezarezvani/claude-skills skill

alirezarezvani/claude-skills is a 345-skill, cross-platform agent skills collection covering engineering, marketing, product, compliance, C-level advisory, research, business operations, commercial finance, and daily productivity. Updated August 6, 2026. Includes 30+ agents, 70+ custom commands, and 330+ skills compatible with Claude Code, Codex, Gemini CLI, Cursor, and 8 additional coding agents. 5,200+ GitHub stars.

What it does for you: Before building any new Axion skill from scratch, checking alirezarezvani/claude-skills surfaces community implementations covering the same pattern — already tested by 5,200+ engineers. The marketing and business operations categories are directly applicable to MCL content and campaign workflows. The C-level advisory and research categories overlap with the scout newsletter and reporting skills Axion already uses.

In practice: cross-platform-skills-library-345-strong

For: Both runtimes. Skill collection compatible with Claude Code, Codex, Gemini CLI, Cursor, and 8 more coding agents — platform-agnostic by the collection's explicit design.

Security3

Quality3

Auditability4

Useful to you4

Useful to community5

Buildable now4

Hermes3

Verdict: watch. 345 skills across all relevant Axion categories; cross-platform compatibility confirmed. Check here before commissioning any new Axion skill build; adopt only after passing the 7-axis quality gate.

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

Source · alirezarezvani/claude-skills GitHub repo confirmed via search results. 345 skills across 15+ categories confirmed. Aug 6 2026 update confirmed. Cross-platform compatibility (Claude Code, Codex, Gemini CLI, Cursor + 8 more) confirmed. 5,200+ stars confirmed via agensi.io research article.

#13

watchBoth runtimes25 / 35

Agno v2.8.6 repo

Agno v2.8.6 (late July through early August 2026) adds two new capability areas over the registry's last version v2.8.4: (1) text-to-speech via Smallest AI's Lightning models — agents can now synthesize audio output directly from the Agno framework; (2) OpenSearch vector database support with vector, keyword, and hybrid search modes. Prior v2.8.x sprint also shipped rollouts (successful runs logged as training data), a durable filesystem layer, and an ops agent that answers plain-English questions about your own agent platform.

What it does for you: TTS via Smallest AI Lightning opens a path for Hermes to produce audio versions of MCL coaching summaries or voice-activate agent outputs — applicable to the MCL audio content workflow. OpenSearch support broadens Axion's vector backend options beyond the current Chroma/FAISS defaults without requiring a full framework change. The ops agent pattern — asking 'what ran last night?' in plain English and getting a structured answer — is directly applicable to Axion Mission Control's monitoring gap, where log reads are currently manual.

In practice: framework-sprint-tts-opensearch-ops-agent

For: Both runtimes. Python framework; runs in Hermes and in Claude Code sessions. v2.8.6 is a material move from registry version v2.8.4 (first_seen 2026-07-29).

Security4

Quality4

Auditability3

Useful to you3

Useful to community4

Buildable now4

Hermes3

Verdict: watch. TTS and OpenSearch are material additions over last registry version v2.8.4; ops agent pattern is directly applicable to Axion monitoring. Stable upgrade path; no breaking changes reported.

Build #13 Agno v2.8.6: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/agno-agi/agno/releases. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · Agno v2.8.6 TTS via Smallest AI Lightning confirmed via agno.com/blog community-roundup-july-2026. OpenSearch vector DB with vector/keyword/hybrid modes confirmed. Ops agent for plain-English platform queries confirmed. agno-agi/agno releases page confirmed. Prior v2.8.4 was first_seen 2026-07-29 in registry.

#14

newwatchBoth runtimes24 / 35

MCP C# SDK v2.0.0 repo

Microsoft's official MCP C# SDK v2.0.0, announced August 2026, implements the 2026-07-28 MCP specification. Headline architecture decisions: stateless-first HTTP (HttpServerTransportOptions.Stateless defaults to true), multi-round-trip requests (MRTR — tools that require client clarification in mid-execution), discovery-first negotiation, caching hints, standardized MCP-Method and MCP-Name HTTP headers, plus two new extension packages (MCP Apps for interactive UI, MCP Tasks for long-running background tasks). Backward compatible: v1 code compiles and runs unchanged.

What it does for you: Microsoft's day-0 implementation of the 2026-07-28 spec in C# confirms the stateless-first HTTP architecture as the production path — not the stateful WebSocket path that legacy MCP servers use. The backward-compatibility guarantee validates that Python MCP servers built on mcp==2.0.0 (yesterday's rank 2) do not need a complete rewrite when clients upgrade. The MRTR pattern — tools that call back to the client for clarification before completing — is the missing piece for Hermes approval flows where a tool needs user confirmation before executing a destructive action.

In practice: spec-validation-reference-impl-mrtr

For: Both runtimes. Official C# reference implementation of the 2026-07-28 MCP spec; architecture signals (stateless-first, MRTR) are directly applicable to Hermes Python MCP server migration strategy.

Security4

Quality5

Auditability4

Useful to you2

Useful to community5

Buildable now2

Hermes2

Verdict: watch. YY is Python-first; C# SDK is not directly importable but Microsoft's day-0 implementation validates the spec architecture and confirms the MRTR pattern for interactive tool approval flows in Hermes.

Build #14 MCP C# SDK v2.0.0: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/modelcontextprotocol/csharp-sdk/releases/tag/v2.0.0. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · modelcontextprotocol/csharp-sdk v2.0.0 release confirmed via devblogs.microsoft.com/dotnet/ announcement. Stateless-by-default HTTP confirmed. MRTR (multi-round-trip requests) confirmed. MCP Apps and MCP Tasks extension packages confirmed. Backward compatibility with v1 code confirmed. InfoWorld coverage of the announcement confirmed.

#15

newwatchStandalone tool24 / 35

Agents-A1 35B MoE (InternScience) repo

Agents-A1 is a 35B-parameter Mixture-of-Experts agentic model from InternScience, open-sourced June 26, 2026 under Apache 2.0. Activates approximately 3B parameters per forward pass (35B-A3B architecture). Claims GAIA 95.1%, BrowseComp 66.8%, IFEval 94.8% — competitive with GPT-5.5 and DeepSeek-V4-Pro on agentic benchmarks. GGUF quantized variants (Q4_K_M, Q8_0) available for local inference. Supports function calling, code interpreters, search engines, and task environments. A 4B variant also released for edge deployment.

What it does for you: GGUF Q4_K_M at 35B fits in approximately 22GB of RAM with llama.cpp — YY's Windows machine would run the 4-bit quantized version without a GPU. For MCL-sensitive coaching content that should not leave the machine, Agents-A1 offers a frontier-class agentic reasoning model with no per-inference API cost. The Apache 2.0 license means it can be embedded in Hermes without royalty concern. The 4B variant could run on lower-memory hardware or edge devices.

In practice: local-frontier-agent-model-gguf-apache2

For: Standalone tool. Open-weights model with GGUF quantizations; runs locally as a standalone inference process via llama.cpp, not a Python library import.

Security4

Quality4

Auditability3

Useful to you3

Useful to community5

Buildable now3

Hermes2

Verdict: watch. GGUF availability makes local inference today-capable on YY's machine; Apache 2.0 is clean. However, the release is 6 weeks old and output quality on MCL coaching-style tasks has not been validated against Claude Sonnet.

Build #15 Agents-A1 35B MoE: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/InternScience/Agents-A1. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · InternScience/Agents-A1 GitHub repo confirmed. Apache 2.0 license confirmed. June 26, 2026 release date confirmed via explainx.ai coverage. GAIA 95.1%, BrowseComp 66.8%, IFEval 94.8% benchmarks confirmed. GGUF Q4_K_M and Q8_0 quantizations confirmed at huggingface.co/InternScience. 35B-A3B MoE architecture confirmed. 4B variant confirmed.

#16

newwatchBoth runtimes24 / 35

tolkonepiu/best-of-mcp-servers repo

best-of-mcp-servers is a weekly-updated ranked list of MCP servers maintained by tolkonepiu, covering 400+ servers across 34 categories updated automatically from GitHub activity metrics each week.

What it does for you: Before building a custom MCP server for any new Axion capability (calendar, email, analytics, monitoring), checking best-of-mcp-servers surfaces existing implementations ready to install. The weekly update cadence and 34-category structure make it faster to find domain-specific servers than a keyword search. Combined with Agent-Leaderboard (rank 10) and skills.sh (rank 2), this completes a three-source pre-build discovery stack.

In practice: weekly-curated-mcp-server-rankings-34-categories

For: Both runtimes. Discovery reference; the MCP servers it lists are usable from both Claude Code and Hermes.

Security3

Quality3

Auditability4

Useful to you3

Useful to community5

Buildable now3

Hermes3

Verdict: watch. Discovery reference for the pre-build MCP server intake step; bookmark and check before building any new Hermes MCP server from scratch.

Build #16 tolkonepiu/best-of-mcp-servers: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/tolkonepiu/best-of-mcp-servers. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · tolkonepiu/best-of-mcp-servers confirmed via search results. 400+ servers across 34 categories confirmed. Weekly update cadence from GitHub activity metrics confirmed.

#17

newwatchClaude (Axion)24 / 35

OpenClaudia/openclaudia-skills skill

openclaudia-skills is a collection of 34 open-source marketing skills for Claude Code covering SEO, content strategy, email marketing, paid ads, analytics, and growth. MIT-licensed. Designed for marketing and content operations workflows in Claude Code sessions.

What it does for you: The MCL content and campaign work YY does in Claude Code sessions could draw on these 34 pre-built marketing skills instead of writing prompts from scratch. The SEO, email, and content strategy skills in particular overlap with MCL's newsletter, content bank, and GHL email flows. As with all community skill collections, each skill requires a 7-axis quality gate before adoption.

In practice: open-source-marketing-skills-mcl-applicable

For: Claude (Axion). Claude Code skills; MIT-licensed collection explicitly targeting Claude Code sessions.

Security4

Quality3

Auditability4

Useful to you3

Useful to community4

Buildable now4

Hermes2

Verdict: watch. 34 marketing skills directly applicable to MCL content operations; MIT-licensed. Check before building new MCL-specific content skills from scratch.

Build #17 OpenClaudia/openclaudia-skills: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/OpenClaudia/openclaudia-skills. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · OpenClaudia/openclaudia-skills GitHub repo confirmed via search. 34 skills covering SEO, content, email, ads, analytics, growth confirmed. MIT license confirmed.

#18

newwatchBoth runtimes22 / 35

Aradotso/ai-agent-skills skill

Aradotso/ai-agent-skills automatically generates agent skills from trending AI agent repositories, published by ara.so. The repo crawls trending GitHub repos and synthesizes skills from their patterns, keeping the skill list current with what the community is actively building. Complementary to the curated directories (skills.sh, best-of-mcp-servers) by surfacing cutting-edge patterns before curation catches up.

What it does for you: For the scout newsletter's pre-build check step, this repo surfaces skills generated from this week's trending repos — patterns that haven't yet made it to the larger curated directories. When the build-intake gate checks skills.sh and alirezarezvani/claude-skills and finds no match, a secondary check here covers the freshest community patterns.

In practice: auto-generated-trending-repo-skills-freshest-patterns

For: Both runtimes. Auto-generated skills from trending agent repos; cross-platform output usable from Claude Code or any MCP-compatible client.

Security3

Quality2

Auditability4

Useful to you3

Useful to community4

Buildable now3

Hermes3

Verdict: watch. Auto-generated skills have higher variance than curated collections; useful as a freshness complement to skills.sh and alirezarezvani/claude-skills, not as a standalone source.

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

Source · Aradotso/ai-agent-skills repo confirmed via search. Auto-generated skills from trending GitHub repos description confirmed. ara.so publisher confirmed.

#19

newwatchBoth runtimes24 / 35

travisvn/awesome-claude-skills skill

awesome-claude-skills is a curated list of Claude Skills, resources, and tools for customizing Claude AI workflows — particularly Claude Code. Maintained by travisvn. Structured as a human-curated awesome-list with descriptions, category tags, and quality annotations rather than an auto-generated index.

What it does for you: For YY's pre-build skill discovery, the human-curated list offers a quality signal that auto-generated indexes lack — entries made it in because a maintainer judged them worthy, not just because a crawler found them trending. Use as a tertiary discovery source after skills.sh and alirezarezvani/claude-skills, especially when looking for rare or niche skill types not well-covered by the larger indexes.

In practice: human-curated-claude-skills-awesome-list

For: Both runtimes. Curated list of Claude Skills resources; the skills it indexes are usable from Claude Code and other MCP-compatible agents.

Security4

Quality3

Auditability5

Useful to you2

Useful to community4

Buildable now4

Hermes2

Verdict: watch. Human-curated quality signal makes this a useful tertiary discovery source; pair with the 7-axis gate before adopting any listed skill.

Build #19 travisvn/awesome-claude-skills: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/travisvn/awesome-claude-skills. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · travisvn/awesome-claude-skills confirmed via search. Human-curated format with descriptions and category tags confirmed. Claude Code workflow focus confirmed.

#20

newwatchBoth runtimes24 / 35

Snyk ToxicSkills security audit repo

Snyk's ToxicSkills audit (February 2026, 3,984 skills analyzed) found that 36% of community Claude Code skills (1,467 skills) had at least one security flaw, 13.4% (534) had critical-severity issues, and 76 were confirmed human-reviewed malicious payloads designed to exfiltrate data or execute arbitrary code. The audit covered prompt injection, credential exposure, unsafe tool calls, and supply-chain risks in skill metadata.

What it does for you: Every time YY or Hermes installs a community skill — from alirezarezvani/claude-skills, openclaudia-skills, awesome-claude-skills, or any other source — there is a 1-in-3 chance of a security flaw and a 1-in-7 chance of a critical one. This finding makes the 7-axis quality gate (particularly the security axis) non-optional for every skill adoption. For Hermes unattended sessions that auto-install or auto-load skills, the risk surface is higher because the malicious payload executes without human review in the loop.

In practice: skill-ecosystem-security-baseline-critical-finding

For: Both runtimes. Security research applicable to any runtime that consumes community skill files — both Claude Code sessions and Hermes tool calls.

Security5

Quality5

Auditability5

Useful to you2

Useful to community5

Buildable now1

Hermes1

Verdict: watch. The 36% flaw rate and 76 confirmed malicious payloads mean community skill adoption requires a security check, not just a usefulness check. No code to install; this is a security posture calibration finding.

Build #20 Snyk ToxicSkills security audit: use the ai-implementation-build-intake skill to build this safely. Source: https://snyk.io/blog/toxicskills-security-audit-open-source-claude-skills/. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · Snyk ToxicSkills audit Feb 2026 confirmed via search. 3,984 skills analyzed confirmed. 36% (1,467) with at least one security flaw confirmed. 13.4% (534) with critical issues confirmed. 76 confirmed human-reviewed malicious payloads confirmed.