Daily AI Implementation Scout Council

2026-08-18. Top pick: #1 bozhouDev/anthropic-mind. Each item is graded on 7 axes; copy a build command to act on it.

Today's ranked top 20

#1
newbuild nowBoth runtimes31 / 35

bozhouDev/anthropic-mind skill

Claude Agent Skill distilled from 428 primary Anthropic sources: blog posts, research papers, podcasts, YouTube transcripts, and X threads. Delivers a single judgment on any Anthropic-shaped trade-off, with a failure condition and a citation back to the source document. Ships 8 worldview models, 21 decision precedents, a flip-case register, and 30 acceptance test questions.

What it does for you: When building AXION agents or designing Hermes behaviours, YY repeatedly faces the question of how Anthropic would handle a specific safety, character, or engineering trade-off. This skill answers that question with a sourced judgment instead of a guess. The full corpus (428 files) travels with the skill, so every conclusion is traceable in the same git clone. The 30 eval questions let YY verify the skill is calibrated before relying on it.

In practice: 4 GitHub stars is an unusually low count for the scope of work. The structured distillation with worldview.md, gates.md, domain references, 21 precedents, and 30 eval tests signals a practitioner who wanted the skill to work correctly, not to go viral. The language is Mandarin Chinese throughout the reference files, with an English SKILL.md entry point.

For: Both runtimes. File-based SKILL.md with a references/ corpus; loads in Claude Code, Codex, Hermes, or any skill-compatible runner with no build step.

Security4
Quality4
Auditability5
Useful to you5
Useful to community4
Buildable now5
Hermes4

Verdict: build now. 31 total. Immediately installable via git clone. The full 428-source corpus is auditable; no external API calls required. Directly answers AXION design questions that currently require YY to re-read Anthropic primary sources each time. Unique: no equivalent Anthropic knowledge skill exists in the registry.

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

Source · 428 primary Anthropic sources indexed as of 2026-08-16 per the repo README corpus cutoff date. Repo structure includes 10 distillation artifacts, 21 decision precedents, 30 acceptance test questions, and 8 worldview models with failure conditions. Corpus spans 32 engineering posts, 170 research papers, 55 YouTube transcripts, and 71 X threads.

#2
newtest firstClaude (Axion)28 / 35

agent-sh/agentsys repo

Modular agent orchestration runtime and plugin marketplace for AI coding environments. Ships 24 plugins, 49 agents (39 file-based plus 10 role-based specialists), and 44 skills that compose into structured development pipelines. Each plugin lives in its own standalone repo under the agent-sh org; agentsys is the marketplace and installer. Pipelines enforce phase gates so agents cannot skip steps. State persists across sessions.

What it does for you: AXION currently assembles agents by hand: one skill here, one hook there, no phase enforcement. agentsys provides a tested runtime that enforces pipeline order, persists state across sessions, and lets YY install any of 24 existing plugins without rebuilding them. The /can-i-help and /onboard agents are proven on real tasks. The benchmark data shows structured prompts via agentsys match raw Opus-tier output using Sonnet-tier cost.

In practice: 967 stars and mentioned in awesome-claude-code. The benchmark numbers (measured March 2026 on real glide-mq tasks with JSON output verification) are the kind of evidence a system deserves before being called production-ready. The 3,518 tests and CI badge confirm the maintainer runs a proper quality gate. Active: pushed 2026-08-17.

For: Claude (Axion). npm package and marketplace installer targeting Claude Code, OpenCode, Codex CLI, Cursor, and Kiro. JavaScript runtime with 30,000 lines of library code and 3,518 tests.

Security3
Quality4
Auditability3
Useful to you5
Useful to community5
Buildable now4
Hermes4

Verdict: test first. 28 total. Strong evidence of quality but the npm install pulls 24 external plugin repos at install time. Each plugin repo must be audited separately before trusting it with AXION data or credentials. Test on an isolated project first.

Build #2 agent-sh/agentsys: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/agent-sh/agentsys. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · 967 GitHub stars as of 2026-08-18. 30,000 lines of library code. 3,518 tests with CI passing. 24 plugins, 49 agents, 44 skills across the agent-sh org. Benchmark measured March 2026 on glide-mq tasks: Sonnet 4 plus agentsys matched raw Opus 4 output quality, verified via claude -p --output-format json.

#3
newtest firstBoth runtimes28 / 35

adand-91/requirement-ledger skill

Retrospective extraction skill for vibe-coded projects. Point it at a project already built with an AI agent: it reads the conversation transcript, surfaces what you actually needed versus what the agent built, identifies recurring pain patterns, and flags repeated work worth converting into a permanent Skill. Bilingual README in English and Simplified Chinese. Created 2026-08-16, topics: agent-skills, claude-code, codex, requirements-engineering.

What it does for you: YY builds AXION tools in vibe-coding sessions and routinely discovers after the fact that the agent solved the wrong problem or repeated a task that deserved to be a skill. requirement-ledger reads those past sessions automatically and produces a structured retrospective: actual requirement, gap, and skill candidates. Run it on any finished AXION build to surface the next skill to write.

In practice: 5 stars on a two-day-old repo. The concept is directly on-point for YY's workflow: the skill reads what the agent said, not just what it produced. The bilingual README and Chinese-language credits suggest the author is active in the same East Asian Claude community that also produced bozhouDev/anthropic-mind and Songzhi-lab/chinese-font-selector this week.

For: Both runtimes. SKILL.md installable in Claude Code or Codex CLI via git clone; reads local conversation transcripts with no external calls.

Security5
Quality3
Auditability4
Useful to you5
Useful to community3
Buildable now5
Hermes3

Verdict: test first. 28 total. Immediately useful for any finished AXION build. Low risk: reads conversation files locally with no network calls. Held at test first because the extraction quality on long multi-session transcripts is unverified; run it on one completed AXION project before treating its output as authoritative.

Build #3 adand-91/requirement-ledger: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/adand-91/requirement-ledger. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · 5 GitHub stars as of 2026-08-18. Created 2026-08-16. Topics: agent-skills, ai-agents, claude-code, codex, developer-tools, llm, prompt-engineering, requirements-engineering. Corpus research/ cutoff stated as 2026-08-16. Git clone install confirmed in README for both Claude Code and Codex CLI.

#4
newtest firstBoth runtimes29 / 35

agent-sh/agnix repo

CLI and LSP linter for agent configuration files: skills, agents, hooks, and MCP tools. 423 validation rules covering frontmatter, tool declarations, permission scopes, and hook wiring. Real-time IDE validation, auto-suggestions, and auto-fix. Catches configuration errors before they fail silently at runtime. Cited in the agentsys README as the companion linter for the agent-sh ecosystem.

What it does for you: AXION has a growing library of skills, hooks, and agent configs that have failed silently in production at least twice (the broken hook that wrote python3 shim noise, the build-gate.js step 0 gap). agnix would catch those errors at write time in the IDE rather than at 2am during a nightly run. 423 rules across 12 platforms means it also validates against the Hermes and Codex variants of the same config files.

In practice: Discovered via the agentsys README. No direct star count or repo confirmation was run in this session; the URL https://github.com/agent-sh/agnix is cited from the README and must be verified on first install. The breadth of 12 platform targets and 423 rules suggests a mature, methodically built tool rather than a quick companion script.

For: Both runtimes. CLI tool and LSP server that validates skill, agent, hook, and MCP config files across Claude Code, Codex, OpenCode, Cursor, Kiro, Copilot, Gemini CLI, Cline, Windsurf, Roo Code, and Amp.

Security5
Quality4
Auditability5
Useful to you4
Useful to community4
Buildable now4
Hermes3

Verdict: test first. 27 total. Read-only linting tool with no credential access and no network calls at validation time. Held at test first because the repo details were not directly verified in this session: confirm the repo exists and review the rule list before wiring it into the AXION IDE setup.

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

Source · 423 validation rules cited in the agentsys README (https://github.com/agent-sh/agentsys). 12 supported platforms listed: Claude Code, Codex, OpenCode, Cursor, Kiro, Copilot, Gemini CLI, Cline, Windsurf, Roo Code, Amp. URL: https://github.com/agent-sh/agnix. Direct repo metrics not confirmed in this session.

#5
newtest firstBoth runtimes27 / 35

Pipelex/pipelex repo

Declarative language for composable AI workflows. A method is a .mthds file specifying typed inputs, typed outputs, and a prompt template. Pipelex handles model routing across 60+ models, structured output parsing, and pipeline orchestration. Methods chain, nest, and share. MIT license. 722 GitHub stars. Pushed 2026-08-17. Has a community hub at mthds.sh for sharing methods.

What it does for you: AXION workflow chains are currently ad-hoc: one agent calls another via raw prompts with no type enforcement and no repeatable execution contract. Pipelex replaces that with a .mthds file per workflow step where every input and output type is declared and every prompt is a versioned template. The model-routing layer means the same method file works on Claude, Gemini, or a local model with no code change. The community hub at mthds.sh gives YY a library of pre-built methods to adapt.

In practice: 722 stars and an active changelog since early 2026. The .mthds format is genuinely readable: a TOML-like file that any team member can understand without knowing Python. The Claude Code integration screenshot in the README shows it working in practice, not just in theory. The mthds.sh hub is a modest but real sign of community momentum.

For: Both runtimes. Python package installable via pip; declarative .mthds files are language-agnostic and work with any model. Claude Code integration shown in repo README screenshot.

Security4
Quality4
Auditability4
Useful to you4
Useful to community4
Buildable now4
Hermes3

Verdict: test first. 27 total. Declarative model routing and typed workflow contracts are a direct improvement over AXION's current ad-hoc chaining. Held at test first because the DSL learning curve and the 60-model routing layer add complexity that needs a real AXION workflow as a proving ground before committing.

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

Source · 722 GitHub stars as of 2026-08-18. MIT license. PyPI package pipelex available. 60+ supported models per README. Community hub mthds.sh live. Repo pushed 2026-08-17. Test badge present in README. Documentation at docs.pipelex.com.

#6
newtest firstStandalone tool25 / 35

triggerdotdev/trigger.dev repo

Open-source platform to build and deploy fully managed AI agents and background workflows. Agents run as durable, resumable tasks with built-in retry, scheduling, and concurrency control. 16,049 GitHub stars. TypeScript SDK. Supports long-running agents that survive serverless timeouts. Can trigger on webhooks, schedules, or events. Self-hostable.

What it does for you: AXION nightly routines are currently fragile: they run as scheduled tasks in Windows Task Scheduler with no retry logic, no distributed state, and no visibility when a step fails silently. trigger.dev would replace that layer: each nightly routine becomes a durable agent task with automatic retry, a dashboard showing run history, and the ability to resume after a network hiccup. The 16k star count and active development confirm it is production-grade infrastructure.

In practice: 16,049 stars and a well-known name in the developer infrastructure community. The AI agent framing is newer (the project started as a background job runner), but the underlying durability guarantees are exactly what AXION nightly automation needs. Self-hostable means YY can run it on the AXION Windows box without a cloud dependency.

For: Standalone tool. Cloud-hosted or self-hosted background job platform with an AI agent SDK. Agents run as managed background tasks, not inside Claude Code or Hermes sessions.

Security3
Quality4
Auditability3
Useful to you4
Useful to community5
Buildable now3
Hermes3

Verdict: test first. 25 total. Production-grade background job infrastructure with AI agent support directly relevant to AXION automation reliability. Held at test first because the self-hosted setup on Windows 11 is undocumented; verify the Windows deployment path before committing AXION routines to it.

Build #6 triggerdotdev/trigger.dev: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/triggerdotdev/trigger.dev. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · 16,049 GitHub stars as of 2026-08-18. Open-source with self-host option. TypeScript SDK. Listed on the GitHub trending workflow-automation search result for today. Active pushes throughout 2026.

#7
test firstStandalone tool26 / 35

langfuse v4.12.0 repo

Open-source LLM observability and evaluation platform. v4.12.0 released 2026-08-17 adds a polished in-app Langfuse Assistant (an AI agent embedded in the Langfuse UI), events-only mode for dataset runs to reduce storage, and trace/session bookmarking removal to simplify the UI. 1,179 stars added since last featured on 2026-08-05.

What it does for you: The Langfuse Assistant in v4.12.0 means YY can now ask the observability UI natural-language questions about AXION agent traces rather than writing filter queries. Combined with the existing Hermes observability integration that was set up via the langfuse-observability plugin (updated 2026-08-17 in claude-plugins-official), v4.12.0 makes the trace review workflow meaningfully faster.

In practice: The in-app agent interface is a genuine UX shift: querying traces by description rather than by filter is the right abstraction for a solo operator who does not want to learn a query language. The langfuse-observability Claude plugin also received a bump today (commit 5b3d4323 to c427cdfc in claude-plugins-official), signalling coordinated release timing.

For: Standalone tool. Self-hosted or cloud observability platform. Integrates with Claude, OpenAI, and any OpenTelemetry-compatible model call. Not a Claude Code or Hermes plugin.

Security4
Quality4
Auditability4
Useful to you4
Useful to community4
Buildable now3
Hermes3

Verdict: test first. 26 total. Re-listed from watch (2026-08-05) because the v4.12.0 in-app agent is a material change. The Langfuse Assistant is the first time the platform offers a conversational interface to trace data. Worth upgrading and testing the assistant against AXION traces.

Build #7 langfuse v4.12.0: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/langfuse/langfuse. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · v4.12.0 released 2026-08-17 per GitHub releases page. langfuse-observability Claude plugin bumped 2026-08-17 (commit 5b3d4323 to c427cdfc in anthropics/claude-plugins-official). Last featured: 2026-08-05 as watch.

#8
newwatchStandalone tool22 / 35

pipeshub-ai/pipeshub-ai repo

Open-source, fully extensible AI context layer that unifies business data from multiple sources into a single knowledge API for AI agents. 3,371 GitHub stars. Pushed 2026-08-17. Designed to be the retrieval and context layer that sits between an agent and scattered business data stores.

What it does for you: AXION agents currently pull context from the YY brain, MCL brain, and flat JSON files on disk. pipeshub-ai would replace the manual context-routing logic with a unified API: one query returns relevant context from all connected sources, ranked and filtered. Relevant if YY wants to connect the Task Hub, Content Pipeline, and Performance Collector data into a single agent-queryable layer.

In practice: 3,371 stars and active push history suggest real adoption beyond the announcing team. The framing as a context layer rather than a RAG framework is a useful distinction: it is infrastructure, not a model wrapper. The extensibility claim needs to be tested against the actual connector list.

For: Standalone tool. Self-hosted or cloud AI context layer. Connects to business data sources and exposes a unified knowledge API. Not a Claude Code skill or Hermes plugin.

Security3
Quality4
Auditability3
Useful to you4
Useful to community4
Buildable now2
Hermes2

Verdict: watch. 22 total. Held at watch because the connector list and actual extensibility need verification against AXION's data sources before committing to the integration. The concept is sound but the setup cost is high without confirmed compatibility with the AXION data formats.

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

Source · 3,371 GitHub stars as of 2026-08-18. Pushed 2026-08-17. Described as open-source and fully extensible. Listed in the agent workflow automation GitHub search results for today.

#9
newwatchStandalone tool22 / 35

skalesapp/skales repo

Personal AI desktop agent for Windows, macOS, Linux, Android, and iOS. Set a goal and the agent works on its own using local or cloud models. Teams feature pairs two desktop agents to collaborate on tasks. MCP support. Local-first with Ollama integration. 1,668 GitHub stars. TypeScript. Pushed 2026-08-17.

What it does for you: The Teams feature (pairing two desktop agent instances) mirrors the AXION orchestrator-worker pattern in a ready-to-run GUI. Relevant for MCL clients who need a personal agent without running their own Claude Code setup. Also useful as a reference architecture for how a local-first MCP-connected agent handles multi-agent coordination.

In practice: 1,668 stars with cross-platform coverage and a Teams feature is a credible consumer AI agent project. The local-first and privacy framing is aligned with YY's AXION principles. The Electron plus React Native stack is familiar territory.

For: Standalone tool. Cross-platform desktop AI agent app (Electron and React Native). Runs as a standalone application, not inside Claude Code or Hermes.

Security4
Quality3
Auditability3
Useful to you3
Useful to community4
Buildable now3
Hermes2

Verdict: watch. 22 total. Consumer-facing desktop agent, not AXION/Hermes infra. Watch for the Teams architecture and as a potential recommendation to MCL clients who want a personal agent. Not a build priority today.

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

Source · 1,668 GitHub stars as of 2026-08-18. TypeScript, Electron and React Native stack. Topics: agentic-ai, ai-agent, mcp, ollama, local-first, windows, macos, linux, ios, android. Pushed 2026-08-17.

#10
newwatchStandalone tool21 / 35

HKUDS/nanobot repo

Ultra-lightweight, open-source, self-hosted personal AI agent framework in Python. WebUI, tools, memory, MCP support, multi-agent coordination, OpenAI-compatible API, Telegram bot integration, and workflow automation. 47,100 GitHub stars. Created 2026-02-01. Pushed 2026-08-17. Topics include openclaw and personal-ai-assistant.

What it does for you: The Telegram bot integration and nightly workflow automation features overlap with AXION's existing bridge and nightly ingest. Nanobot's self-hosted WebUI could serve as a lightweight alternative to the here.now-hosted tools for team members who need a local agent interface without a public URL. The 47k star count confirms it is battle-tested by a very large user base.

In practice: 47,100 stars is the largest raw count in today's issue. The project has been running since February 2026 and the star velocity suggests it became the default self-hosted personal agent for a large open-source community. The openclaw topic reference connects it to a framework that appeared in the prior issue as a watch item.

For: Standalone tool. Self-hosted Python framework with WebUI. Runs as a standalone server, not as a Claude Code or Hermes component.

Security3
Quality4
Auditability3
Useful to you3
Useful to community5
Buildable now3
Hermes0

Verdict: watch. 21 total. Consumer-facing personal assistant framework, not AXION/Hermes building-block infra. The Telegram bot and MCP layers overlap with AXION but the full self-hosted stack would duplicate rather than extend what already exists. Watch for architectural patterns and as a fallback recommendation.

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

Source · 47,100 GitHub stars as of 2026-08-18. Created 2026-02-01. Python. Topics: agent-framework, ai-agent, mcp, telegram-bot-ai-assistant, self-hosted, workflow-automation, openclaw. Pushed 2026-08-17.

#11
newwatchClaude (Axion)21 / 35

Claude Code 2.1.234 repo

Claude Code 2.1.234 released. Key changes: auto-continue at usage limit (sessions resume automatically when a claude.ai reset fires, configurable in /config); todo/task-tracking tools (TaskCreate, TaskGet, TaskUpdate, TaskList, TodoWrite) removed from Opus 4.8, Sonnet 5, Fable 5, Mythos 5, and newer models (restore with CLAUDE_CODE_ENABLE_TODO_TOOLS=1); NT-namespace path rejection in remote file reads and CLAUDE.md includes (security hardening); GitLab MR badge in footer and statusline; Remote Control effort picks now sync from phone to terminal; credential masking on permission previews now covers shell delimiters.

What it does for you: Auto-continue at usage limit removes the most common reason AXION nightly runs stall: a claude.ai rate limit hitting mid-task. The todo tool removal is an immediate action item: AXION workflows that call TaskCreate or TaskUpdate in unattended sessions will silently fail on Fable 5 and newer models unless CLAUDE_CODE_ENABLE_TODO_TOOLS=1 is set. The NT-namespace path hardening closes the same NTLM credential-leak vector that appeared in 2.1.232 and 2.1.233.

In practice: The todo tool deprecation is the most operationally significant change for AXION. The gentle reminder in this session's system prompt about TaskCreate and TaskUpdate arriving while working means the task reminder tool is still wired in the harness. After this update, using those tools in unattended Fable 5 sessions will produce silent no-ops without the environment variable.

For: Claude (Axion). Claude Code version update. Changes apply to the Claude Code runtime that AXION sessions run on.

Security4
Quality4
Auditability4
Useful to you4
Useful to community4
Buildable now1
Hermes0

Verdict: watch. 21 total. The auto-continue feature and todo tool removal are both actionable today. Watch rather than build-now because no new capability to build exists here; the action is to set CLAUDE_CODE_ENABLE_TODO_TOOLS=1 in the AXION run-bridge environment if unattended task tracking is needed.

Build #11 Claude Code 2.1.234: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/anthropics/claude-code/releases/tag/v2.1.234. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Claude Code 2.1.234 confirmed in CHANGELOG.md as of 2026-08-18. Previous version 2.1.233 confirmed as separate release with GitLab merge request URL support and memory cgroup. Todo/task tool removal listed explicitly for Opus 4.8, Sonnet 5, Fable 5, Mythos 5, and newer. Environment variable CLAUDE_CODE_ENABLE_TODO_TOOLS=1 documented as the restore path.

#12
newwatchStandalone tool20 / 35

keinsaasforever/better-chatbot repo

Navigator (formerly Better Chatbot): open-source AI workspace for agents, MCP, and multi-model conversations. 1,164 GitHub stars. Provides a browser-based interface for connecting multiple AI agents and MCP servers in a unified workspace with persistent sessions.

What it does for you: If MCL team members need a browser-based interface to interact with Hermes agents and AXION MCP servers without installing Claude Code, Navigator provides that layer. Relevant as a lightweight alternative to the here.now-hosted Task Hub for external collaborators who need agent access.

In practice: 1,164 stars and a rename from Better Chatbot to Navigator suggests the project matured past the chatbot framing into a broader agent workspace. The MCP integration is the differentiating factor from generic chat UIs.

For: Standalone tool. Open-source web workspace application. Runs as a self-hosted web server, not as a Claude Code or Hermes component.

Security3
Quality3
Auditability3
Useful to you3
Useful to community4
Buildable now2
Hermes2

Verdict: watch. 20 total. Useful as a browser-based agent workspace but adds infrastructure that overlaps with the existing here.now tooling. Watch for use cases where MCL collaborators need agent access without a local Claude Code install.

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

Source · 1,164 GitHub stars as of 2026-08-18. Formerly named Better Chatbot per the README. MCP server integration listed as a core feature.

#13
newwatchStandalone tool19 / 35

hexabot-ai/Hexabot repo

Hexabot v3: AI workflow automation platform combining visual workflow builder, actions, and multi-channel messaging. 1,179 GitHub stars. Supports agent orchestration and webhook-triggered workflows. Self-hostable.

What it does for you: Hexabot's multi-channel messaging and webhook-triggered workflows could replace the AXION Telegram bridge's manual wiring for specific MCL notification flows. The visual workflow builder lowers the barrier for MCL team members to define automation without touching Python scripts.

In practice: 1,179 stars and a v3 milestone suggest the project has a real user base. The visual builder for non-technical team members is a genuine differentiator from code-first automation tools.

For: Standalone tool. Self-hosted AI workflow automation platform. Runs as a web service, not inside Claude Code or Hermes.

Security3
Quality3
Auditability3
Useful to you3
Useful to community3
Buildable now2
Hermes2

Verdict: watch. 19 total. Interesting as a no-code automation layer for MCL team members but overlaps substantially with the existing AXION bridge and nightly routine system. Watch for the MCL team enablement angle.

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

Source · 1,179 GitHub stars as of 2026-08-18. v3 release confirmed. Listed in the agent workflow automation GitHub search results for today.

#14
newwatchStandalone tool19 / 35

bytechefhq/bytechef repo

Open-source platform unifying AI agent orchestration and workflow automation. 983 GitHub stars. Provides a combined interface for building agents and integrating them into business workflows with pre-built connectors.

What it does for you: Potential replacement for the piecemeal AXION automation stack (Task Scheduler plus Python scripts plus Telegram bridge) with a unified platform that handles scheduling, retries, and agent orchestration in one place.

In practice: 983 stars is solid for a unified orchestration platform. The combination of agents and workflow automation in one tool is appealing but the integration cost is high.

For: Standalone tool. Self-hosted or cloud platform for unified AI agent orchestration and workflow automation. Runs as a separate service.

Security3
Quality3
Auditability3
Useful to you3
Useful to community3
Buildable now2
Hermes2

Verdict: watch. 19 total. Held at watch because it competes with trigger.dev (rank 6) which has 16x the stars and a more proven track record. Evaluate trigger.dev first; if it does not meet the need, revisit bytechef.

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

Source · 983 GitHub stars as of 2026-08-18. Listed in the agent workflow automation GitHub search results for today. Self-host option confirmed in the description.

#15
newwatchBoth runtimes26 / 35

Songzhi-lab/chinese-font-selector skill

Chinese font selection agent skill: 3-tier commercial licensing matrix for Chinese fonts, a scene-by-mood font selection matrix, Chinese-English mixed typesetting rules, and pairing recommendations. 20 GitHub stars. Topics: agent-skills, chinese-fonts, chinese-typography, cjk-fonts, claude-skill, design-system, font-pairing, free-commercial-fonts, skill-md, typography-tools.

What it does for you: When creating MCL social graphics, carousel slides, or PDF deliverables with Chinese text, YY currently picks fonts by feel. This skill gives an agent a structured decision: font name, licensing tier (free commercial, attribution required, or paid), and typesetting rule for the specific scene. Directly relevant for MCL's Chinese-speaking Singapore and Malaysia audience.

In practice: 20 stars on a niche-but-correct skill. The 3-tier licensing classification is practically useful: agents picking commercial-use fonts without licensing guidance can create IP liability. The scene-by-mood matrix suggests someone who has done this work repeatedly and codified the decision tree.

For: Both runtimes. File-based Claude skill (SKILL.md) with a corpus of Chinese font licensing tiers and typography rules. Loads in Claude Code or Hermes with no build step.

Security5
Quality3
Auditability4
Useful to you3
Useful to community2
Buildable now5
Hermes4

Verdict: watch. 18 total. Install and keep for the next MCL Chinese design job. No build required; git clone takes 30 seconds. Low priority today but high value when the specific need arises.

Build #15 Songzhi-lab/chinese-font-selector: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/Songzhi-lab/chinese-font-selector. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · 20 GitHub stars as of 2026-08-18. 12 descriptive topics including font-pairing, free-commercial-fonts, and chinese-typography. SKILL.md format confirmed by skill-md topic tag.

#16
newwatchStandalone tool18 / 35

Pinvou/pinvou-agent repo

Open-source desktop AI agent for tools, files, knowledge, workflows, and real device control. 900 GitHub stars. Cross-platform. Goals-based agent that works autonomously on the local machine.

What it does for you: Reference architecture for a local-first desktop agent with file and tool access. Comparable to skales (rank 9) but with different emphasis on real device control.

In practice: 900 stars and goals-based autonomous operation suggests it is aimed at the same power-user segment as skales. The real device control angle is less relevant for AXION than the knowledge and workflow angles.

For: Standalone tool. Open-source desktop AI agent application. Runs as a standalone desktop app, not as a Claude Code or Hermes component.

Security3
Quality3
Auditability3
Useful to you3
Useful to community3
Buildable now3
Hermes0

Verdict: watch. 18 total. Positioned similarly to skales (rank 9). Watch as a reference for the desktop agent architecture. No action needed today.

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

Source · 900 GitHub stars as of 2026-08-18. Listed in agent workflow automation search results. Goals-based autonomous operation described in the repo description.

#17
newwatchBoth runtimes26 / 35

orange4664/chinese-grammar-proofreader skill

Chinese grammar proofreader Claude skill. Focuses on making AI-written Chinese readable to a human on first pass, not on removing AI tone. Corrects structural sentence errors specific to Chinese grammar. 12 GitHub stars. Topics: chinese, claude-skill, grammar, proofreading, writing-tools.

What it does for you: MCL content produced by Claude for Chinese-speaking audiences often has structural sentence issues that are invisible to non-native readers but obvious to Singaporean or Malaysian Chinese speakers. This skill applies a structured grammar pass aimed specifically at those errors. Pair it with speak-marc for Chinese MCL content that passes a readability check before delivery.

In practice: 12 stars and a focused, single-purpose description. The explicit goal (readable on first pass, not AI-tone removal) is the right priority for MCL content: Marc's voice should sound natural, not just non-AI.

For: Both runtimes. File-based Claude skill (SKILL.md) for Chinese grammar correction. Loads in Claude Code or Hermes with no build step.

Security5
Quality3
Auditability4
Useful to you3
Useful to community2
Buildable now5
Hermes4

Verdict: watch. 17 total. Install and pair with speak-marc on the next Chinese MCL content run. Low priority because the use case is narrow but zero setup cost makes it worth having.

Build #17 orange4664/chinese-grammar-proofreader: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/orange4664/chinese-grammar-proofreader. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · 12 GitHub stars as of 2026-08-18. Topics: chinese, claude-skill, grammar, proofreading, writing-tools. SKILL.md format confirmed by claude-skill topic tag.

#18
newwatchClaude (Axion)21 / 35

Lionad-Morotar/polaris-flow skill

Claude Code skills collection for long-horizon tasks: development workflow, code review, knowledge base management, image generation, UI restoration, and teaching tour. 4 GitHub stars. Designed for models with large context windows and extended task horizons.

What it does for you: The knowledge base management and teaching tour skills are the most relevant for AXION. If the knowledge base skill covers a different indexing pattern than context-mode, it could complement or extend the YY brain workflow.

In practice: 4 stars on a broad skills collection. The individual skill quality is unknown without reading each SKILL.md. The breadth is a risk factor: generalist collections often have thin coverage on each topic.

For: Claude (Axion). Claude Code skills collection explicitly targeting long-horizon capable models. The repo description names Claude Code as the primary runtime.

Security4
Quality2
Auditability3
Useful to you3
Useful to community2
Buildable now4
Hermes3

Verdict: watch. 16 total. Broad collection with no star validation yet. Read each skill file before installing any of them into the AXION canonical library. The teaching tour skill in particular is worth inspecting.

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

Source · 4 GitHub stars as of 2026-08-18. Collection covers: development workflow, code review, knowledge base, image generation, UI restoration, teaching tour per the repo description. Explicitly targets long-horizon capable models.

#19
newwatchBoth runtimes17 / 35

fuqiuluo/ida-headless-mcp plugin

Rust-native multi-session headless IDA Pro MCP server. One supervisor process manages one or more IDA Pro instances. Exposes IDA Pro analysis capabilities (disassembly, breakpoints, memory inspection) to any MCP client. 10 GitHub stars since 2026-08-17.

What it does for you: If AXION ever expands into firmware analysis or binary security work, this MCP server gives any Claude Code session direct IDA Pro access via MCP. The multi-session supervisor pattern (one MCP endpoint, many IDA instances) is architecturally interesting as a model for managing multiple expensive tool instances behind a single MCP interface.

In practice: 10 stars on a very specific security tooling integration. The Rust implementation and multi-session supervisor architecture suggest a practitioner who needed this to work reliably, not just to demo well. Niche but technically sound.

For: Both runtimes. MCP server installable in any MCP-compatible client including Claude Code and Hermes. The supervisor model handles multiple IDA Pro sessions via a single MCP endpoint.

Security3
Quality3
Auditability3
Useful to you1
Useful to community2
Buildable now2
Hermes3

Verdict: watch. 15 total. YY does not currently use IDA Pro, which removes the primary use case. The supervisor MCP pattern is worth studying as an architectural reference. No action needed today.

Install: Build from source with `cargo build --release`, then register as an MCP server in ~/.claude/settings.json under mcpServers. Requires IDA Pro license. Review Rust source before enabling.

Source · 10 GitHub stars as of 2026-08-18. Rust implementation. Created and pushed 2026-08-17. Description: multi-session headless IDA Pro MCP server with supervisor model.

#20
newwatchClaude (Axion)17 / 35

claude-security plugin v0.10.1-rc7 plugin

Official Claude Code security plugin, updated to v0.10.1-rc7 on 2026-08-17. Provides security analysis skills, agents, hooks, workflows, and scripts for Claude Code sessions. Maintained by Anthropic in the official plugins directory (33,600 stars, 3,297 commits).

What it does for you: Adds a Anthropic-maintained security analysis layer to every Claude Code session: skills for vulnerability detection, agents for security review pipelines, and hooks that intercept potentially dangerous commands. The v0.10.1-rc7 release candidate signals a stable v0.10.1 is imminent. Pairs with the bar181/bar-observatory security monitoring approach that appeared in a prior issue.

In practice: RC7 on a minor version suggests the team is being careful about the stable release. The presence of hooks and workflows (not just skills) means this plugin actively monitors sessions, not just answers security questions. The 2026-08-17 commit is the most recent activity across all official plugins today.

For: Claude (Axion). Official Anthropic-managed Claude Code plugin. Installs into Claude Code via the official plugin directory. Contains skills, agents, hooks, workflows, and scripts.

Security2
Quality4
Auditability3
Useful to you3
Useful to community3
Buildable now1
Hermes1

Verdict: watch. 15 total. Anthropic-maintained security plugin with hooks means it runs code on every session. Worth installing on the AXION box once the stable v0.10.1 lands; wait for RC to clear before enabling in unattended nightly runs.

Install: Enable by adding claude-security to enabledPlugins in ~/.claude/settings.json, or run `claude plugin install claude-security` from the anthropics/claude-plugins-official directory. Wait for stable v0.10.1 before enabling in unattended sessions. Review before enabling.

Source · v0.10.1-rc7 commit pushed 2026-08-17 to anthropics/claude-plugins-official (33,600 stars, commit #5403 of 3,297+ total). Plugin directory contains: .claude-plugin, agents, hooks, scripts, skills, and workflows folders per the GitHub tree view.