Daily AI Implementation Scout Council

Today's ranked top 20

#1

newbuild nowBoth runtimes30 / 35

Leutenegger/book-to-skill skill

Python tool that converts technical book PDFs into Claude Code skills, ready to study, reference, and invoke while working. 1,152 GitHub stars and 141 forks in 4 days since August 13, 2026. Parses PDF structure, extracts concepts, and outputs a SKILL.md that an agent can load and query mid-session.

What it does for you: YY's AXION skill library currently requires writing each skill by hand. book-to-skill converts any technical PDF, architecture book, or methodology document directly into a queryable Claude Code skill without manual extraction. Run it on any API documentation, framework guide, or book PDF and get a structured, searchable skill the next day. Directly extends the AXION canonical skill library without authoring time.

In practice: 1,152 stars in 4 days is an extraordinary signal for a single-purpose developer tool. Topics span agent-skills, claude-code-plugin, pdf-parser, and skill-generator, suggesting intentional design rather than an experimental one-day project. The creator cites daily use in their own workflow.

For: Both runtimes. Python tool invokable from any agent context; outputs a SKILL.md that loads in Claude Code, Codex, GitHub Copilot, and any skill-compatible runner.

Security4

Quality5

Auditability4

Useful to you5

Useful to community5

Buildable now4

Hermes3

Verdict: build now. 1,152 stars and 141 forks in 4 days confirms strong community validation. Directly extends AXION skill ecosystem without manual writing. Local Python, auditable, no external API calls required by the core tool.

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

Source · 1,152 GitHub stars and 141 forks as of August 17, 2026; created August 13, updated August 16. Python implementation confirmed. Topics include claude-code-plugin, agent-skills, pdf-parser, pdf-tools, skill-generator. Four days of growth at this star rate is the strongest community signal in today's issue.

#2

newbuild nowBoth runtimes29 / 35

machina-exm/film-studio-skills skill

7 installable agent skills covering the AI video production pipeline from script to locked, generation-ready shot prompts. Integrates with Higgsfield and Seedance for generation. 27 GitHub stars and 3 forks since August 14, 2026. Topics: agent-skills, ai-video, claude-code, higgsfield, seedance.

What it does for you: YY already uses Higgsfield for AI video production at MCL, and the vox-collage-craft skill handles narrated-video workflows. These 7 skills cover the pipeline between the idea and the generation-ready prompt: script development, shot design, and prompt locking. Installing them into Hermes connects the MCL content workflow end to end without manual prompt engineering for each video, and the Higgsfield topic tag confirms the integration targets the tool YY is already paying for.

In practice: The authors describe this as the working pipeline behind multi-million dollar AI film productions condensed into 7 installable skills. The Higgsfield and Seedance topic tags confirm production tool integration rather than a toy example. Low star count at launch is expected for specialist tooling aimed at a narrow professional audience; the Hermes and OpenCode compatibility signals intentional cross-agent design.

For: Both runtimes. Repository explicitly lists Claude Code, Codex, Hermes, and OpenCode as supported runtimes. Install via the agent skills protocol with no build step.

Security4

Quality3

Auditability4

Useful to you5

Useful to community3

Buildable now5

Hermes5

Verdict: build now. Directly extends YY's existing Higgsfield video production workflow. 7 installable skills with confirmed Hermes runtime compatibility. Immediate install, no build required. Production-claimed provenance from AI video pipeline authors.

Build #2 machina-exm/film-studio-skills: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/machina-exm/film-studio-skills. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · 27 GitHub stars and 3 forks as of August 17, 2026. Created August 14, updated August 16. Topics: agent-skills, ai-video, claude-code, higgsfield, seedance. Repository describes 7 discrete installable skills. Hermes listed as a supported runtime in the repository.

#3

newtest firstClaude (Axion)26 / 35

Endokelp/Endoplexity plugin

Agentic browser control Chrome extension and MCP server that uses the Claude or Cursor subscription the user already has, rather than requiring a separate API key. 224 GitHub stars since August 14, 2026. TypeScript, CDP-based browser automation, compatible with Claude Code and Cursor as the controlling agent.

What it does for you: YY already pays for Claude. Endoplexity routes browser automation through that subscription rather than requiring a separate API key or billing account. Compared to vercel-labs/agent-browser (yesterday's top pick, which requires building from Rust source), Endoplexity installs as a Chrome extension: lower setup cost, immediate operability on Windows 11. Worth testing alongside agent-browser to see which fits AXION's automation pattern before committing to either.

In practice: 224 stars in 3 days from a single-author repo is a strong organic signal. The Chrome extension approach is creative: instead of standing up a headless browser service, it rides the existing interactive Claude session. CDP-based automation is the same foundation Playwright uses internally, which is a credible technical choice.

For: Claude (Axion). Chrome extension and MCP server that routes browser control through the Claude or Cursor subscription already active in the user's browser. Requires Chrome and an active Claude subscription.

Security3

Quality4

Auditability3

Useful to you4

Useful to community4

Buildable now4

Hermes4

Verdict: test first. 224 stars and fast traction confirm interest. Held at test first because vercel-labs/agent-browser (stronger community signal, Rust production build, Vercel Labs provenance) was the BUILD NOW pick yesterday and covers the same use case. Evaluate both before choosing AXION's browser automation path.

Install: clone the repo, load the unpacked Chrome extension in developer mode, then add the MCP server to Claude Code config per the README. Not currently in YY's settings.json enabledPlugins list. Review before enabling. Copy

Source · 224 GitHub stars and 0 forks as of August 17, 2026. Created August 14, updated August 16. TypeScript. Topics: ai-agent, browser-automation, cdp, chrome-extension, claude, cursor, mcp. Repository description confirmed.

#4

newtest firstBoth runtimes28 / 35

fellowgeek/mcp-memory plugin

MCP server providing persistent, long-term memory with SQLite FTS5 full-text search for AI agents. OKF-backed indexing approach. 175 GitHub stars and 11 forks since August 13, 2026. Python implementation. Compatible with Claude Desktop, Cursor, Antigravity, Windsurf, and Codex.

What it does for you: AXION agents currently lose memory between sessions unless YY has explicitly wired a brain page. mcp-memory gives any MCP-connected agent a persistent, searchable fact store backed by local SQLite with no cloud dependency. AXION bridge agents running overnight could write facts, decisions, and observations that carry forward to the next session automatically without requiring a manual brain ingest step.

In practice: OKF (Open Knowledge Foundation) backing adds credibility above a standard weekend project. SQLite FTS5 is a well-understood, auditable storage layer: no opaque vector database, no external service. 175 stars in 4 days on a Python MCP server is a strong validation signal for a focused infrastructure tool.

For: Both runtimes. MCP server compatible with Claude Desktop, Cursor, Antigravity, Windsurf, and Codex. Any MCP-enabled agent can connect via standard config.

Security4

Quality4

Auditability4

Useful to you4

Useful to community4

Buildable now4

Hermes4

Verdict: test first. 175 stars, OKF-backed, SQLite FTS5 is auditable local storage. Held at test first because rohitg00/agentmemory (yesterday's issue, TEST FIRST, 92 percent token reduction claim) targets the same problem and has not yet been evaluated. Run both on the same AXION session benchmark before picking AXION's memory layer.

Install: pip install and add MCP server to Claude Code config per README. Not currently in YY's settings.json enabledPlugins list. Review before enabling. Copy

Source · 175 GitHub stars and 11 forks as of August 17, 2026. Created August 13, updated August 16. Python implementation. OKF-backed confirmed. SQLite FTS5 storage and compatible agent list (Claude Desktop, Cursor, Antigravity, Windsurf, Codex) confirmed in README.

#5

newtest firstBoth runtimes28 / 35

nduc99911/repo-context-mcp plugin

MCP server that produces repo maps, code search results, and token-aware context packs for AI coding agents. TypeScript. 103 GitHub stars updated approximately August 13, 2026. Compatible with Claude Code, Cursor, and Codex.

What it does for you: AXION is a 4.7 GB workspace that agents frequently navigate poorly, loading entire files when they need one function. repo-context-mcp gives any agent a structured repo map and token-capped code search results so it reads only what it needs. This directly reduces context waste on every AXION coding session where the agent explores before editing, and the MCL junction and graphify outputs make scoped context even more important here.

In practice: 103 stars on a focused MCP utility for a problem every Claude Code user encounters is a credible signal. TypeScript with MCP protocol compliance suggests it will integrate without friction. The token-aware context packing is the differentiating feature: not just search, but search with a token budget enforced.

For: Both runtimes. TypeScript MCP server compatible with Claude Code, Cursor, and Codex via standard MCP config. Works with any MCP client that follows the 2026-07-28 spec.

Security4

Quality4

Auditability4

Useful to you4

Useful to community4

Buildable now4

Hermes4

Verdict: test first. 103 stars, fills a real daily friction point in AXION: agent context waste from poor repo navigation. Immediate install via MCP config with no build step beyond npm.

Install: npm install and add the MCP server to Claude Code config per README. Not currently in YY's settings.json enabledPlugins list. Review before enabling. Copy

Source · 103 GitHub stars as of August 17, 2026. TypeScript implementation. Topics: typescript, mcp, cursor, codex, claude. Updated approximately August 13. Repository description confirms repo map, code search, and token-aware context packs.

#6

newtest firstBoth runtimes27 / 35

idavidov13/agentic-playwright repo

Production-grade Playwright and TypeScript scaffold for agentic end-to-end testing, with harness configuration for all major AI coding agents pre-wired. 90 GitHub stars and 25 forks since August 13, 2026. Python repo with TypeScript test surface. Topics include claude-code, cursor, e2e, playwright, test-automation.

What it does for you: AXION has growing web surface area across here.now sites, the Content Pipeline Hub, and bridge tools. agentic-playwright gives Claude Code a ready-made scaffold to generate, run, and fix Playwright E2E tests without the usual boilerplate setup. Point a Claude Code session at any AXION web tool and ask it to write tests immediately, using the harness configuration already in the scaffold.

In practice: Production-grade claim is backed by AGENTS.md harness integration rather than just comments in code. 90 stars in 4 days for a testing scaffold is notable since testing tools typically attract fewer stars than feature tools. The 25 forks suggest developers are already forking and adapting it.

For: Both runtimes. AGENTS.md and harness configuration baked in for Claude Code, Cursor, GitHub Copilot, and Codex. Works with any AI coding agent that reads AGENTS.md.

Security4

Quality4

Auditability4

Useful to you4

Useful to community4

Buildable now4

Hermes3

Verdict: test first. 90 stars, AGENTS.md harness works with Claude Code, directly addresses AXION's under-tested web surfaces. One integration attempt on a real AXION site before recommending fully.

Build #6 idavidov13/agentic-playwright: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/idavidov13/agentic-playwright. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · 90 GitHub stars and 25 forks as of August 17, 2026. Created August 13, updated August 16. Topics: testing, typescript, ai, api-testing, claude-code, cursor, e2e, github-copilot, playwright, test-automation. Claude Code harness support confirmed in topic list.

#7

newtest firstClaude (Axion)27 / 35

bar181/bar-observatory plugin

Deterministic, local-only flight recorder and audit tool for Claude Code AI agent sessions. Rust and SQLite. Zero network calls. Reads every tool call, task, sub-agent dispatch, rework loop, and error from the session and stores it locally as structured data. MCP server interface exposes the audit data back to agents during sessions. 17 GitHub stars since August 12, 2026.

What it does for you: AXION runs dozens of scheduled Claude Code sessions overnight (daily-ingest, AI scout, friction-audit, Marc digest). The only current record is the runs.log heartbeat. bar-observatory gives YY a structured, queryable audit trail for every session: tool calls per task, sub-agent dispatches, rework loop counts, and error locations. Directly useful for diagnosing overnight session failures without reading raw transcripts.

In practice: Local-first, Rust, SQLite, zero network calls is a strong design philosophy for a monitoring tool that itself must be trustworthy. The MCP server pattern means an agent can query its own audit history during a session, not just after. Low star count for a young, specialist tool is expected; the technical depth and clear scope are the signal.

For: Claude (Axion). Claude Code specific: reads Claude Code session transcripts and exposes audit data back to Claude Code via MCP. Cannot audit other runtimes without significant modifications.

Security5

Quality3

Auditability5

Useful to you4

Useful to community3

Buildable now4

Hermes3

Verdict: test first. Fills a real AXION monitoring gap: structured audit trails for overnight Claude Code sessions. Rust and SQLite with zero network calls is auditable and safe. Held at test first rather than build now due to low star count (17) and no confirmed external users beyond the author.

Install: compile from Rust source (cargo build --release) or watch for a release binary, then add MCP server to Claude Code config. Not currently in YY's settings.json enabledPlugins list. Review before enabling. Copy

Source · 17 GitHub stars as of August 17, 2026. Created August 12, updated August 15. Rust and SQLite implementation confirmed. Zero network calls stated explicitly in README. Topics: agent-monitoring, claude-code, claude-code-plugin, deterministic, llm-observability, local-first, mcp, mcp-server, rust, sqlite, telemetry.

#8

newtest firstBoth runtimes25 / 35

ZSeven-W/dsh-noema plugin

Noema long-term memory plugin for DeepSeek Harness. Provides durable, inspectable agent memory backed by the Noema cloud memory service. Features: recall-before-work (auto-loads relevant memories before a task starts), import from 9 agent tools, settings-page memory management, crash keep-alive, hot reload. TypeScript. 88 GitHub stars and 7 forks since August 14, 2026.

What it does for you: If YY evaluates DeepSeek Harness as a cross-agent orchestration layer (dsh-plugin-dev-skills appeared yesterday), dsh-noema gives it persistent memory with a visual settings page and automatic recall before each task. The import-from-9-tools feature could absorb memory from existing AXION tools. Worth testing alongside mcp-memory and agentmemory to see which memory approach handles multi-session AXION workflows best.

In practice: 88 stars in 3 days on a DSH plugin with a commercial Noema backend is a real signal. The recall-before-work feature is a concrete UX improvement over generic memory stores: the agent loads relevant memories automatically rather than requiring explicit recall commands. TypeScript and crash keep-alive suggest production-minded design.

For: Both runtimes. DeepSeek Harness plugin with MCP compatibility. Primary use is as a DSH plugin; secondary use as a standalone MCP server for non-DSH agents.

Security4

Quality3

Auditability4

Useful to you3

Useful to community4

Buildable now4

Hermes3

Verdict: test first. 88 stars and solid feature set for agent memory. Held at test first because it depends on the Noema cloud service (cost and terms unconfirmed) and requires evaluating the DSH ecosystem alongside it.

Install: npm install the DSH plugin or configure as MCP server per README. Requires a Noema account (cost unconfirmed). Not currently in YY's settings.json enabledPlugins list. Review before enabling. Copy

Source · 88 GitHub stars and 7 forks as of August 17, 2026. Created August 14, updated August 16. TypeScript. Topics: agent-memory, deepseek-harness, dsh-plugin, long-term-memory, mcp, noema. Recall-before-work, 9-tool import, crash keep-alive features confirmed in README.

#9

newtest firstBoth runtimes25 / 35

memoket/memoket-kite repo

Memory engine for AI agents with token-efficient, explainable retrieval beyond vector similarity. Python. 65 GitHub stars and 5 forks since August 12, 2026. Combines semantic, keyword, and entity matching to produce retrieval results with explanations of why each memory matched, not just what matched.

What it does for you: Standard vector memory stores give agents answers without audit trails. memoket-kite returns retrieval results with explanations, which makes debugging agent memory failures much easier. For AXION sessions where agents sometimes retrieve wrong context and produce wrong outputs, this explainability could halve the debugging time on memory-related failures.

In practice: 65 stars in 5 days for a memory library with a specific explainability claim is a focused signal. The beyond-vector-similarity positioning addresses a genuine criticism of RAG-style memory: it works but you cannot audit why. Python implementation keeps it accessible.

For: Both runtimes. Python library that any agent pipeline can import. Integrates with standard retrieval patterns so any agent framework can call it directly.

Security4

Quality3

Auditability4

Useful to you3

Useful to community4

Buildable now4

Hermes3

Verdict: test first. 65 stars, explainable retrieval addresses a real AXION debugging pain point. Held at test first because three memory tools appeared this week (memoket-kite, mcp-memory, dsh-noema); evaluate all three on the same benchmark before picking one.

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

Source · 65 GitHub stars and 5 forks as of August 17, 2026. Created August 12, updated August 16. Python. Topics: agent-memory, ai-memory, llm-memory, long-term-memory, memory-engine, rag, retrieval. Explainable retrieval claim confirmed in repository description.

#10

newwatchBoth runtimes25 / 35

danyuchn/iso-24495-skill skill

ISO 24495-1 plain language standard implemented as a Claude Code skill. Bilingual: English and Traditional Chinese. Reader-outcome rewriting: makes text relevant, findable, understandable, and usable. 105 GitHub stars and 3 forks since August 13, 2026.

What it does for you: MCL produces high volumes of written content for courses, SOPs, and client materials. Applying ISO 24495-1 plain language rules ensures content is readable at the intended level without manual editing passes. The Traditional Chinese bilingual capability covers YY's Chinese-language audience directly.

In practice: 105 stars on a skill implementing a specific ISO standard is a niche but genuine signal. The bilingual English and Traditional Chinese design shows intentional regional coverage. Instant install as a Claude Code skill means zero setup cost and no dependencies.

For: Both runtimes. Pure SKILL.md text file, no code execution. Works in Claude Code, Codex, and any skill-compatible runner.

Security5

Quality3

Auditability5

Useful to you2

Useful to community2

Buildable now5

Hermes3

Verdict: watch. 105 stars and useful for MCL content quality. Numerically scores at the test-first threshold but held at WATCH because content-quality skills are lower priority than agent infrastructure, and ISO 24495-1 rules work better after a one-time evaluation session with a real MCL document before integrating into production workflows.

Build #10 danyuchn/iso-24495-skill: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/danyuchn/iso-24495-skill. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · 105 GitHub stars and 3 forks as of August 17, 2026. Created August 13, updated August 16. Bilingual English and Traditional Chinese confirmed in description. ISO 24495-1 standard scope confirmed. Zero code execution: pure SKILL.md text.

#11

newwatchClaude (Axion)23 / 35

marvin-socialista/claude-design-mcp plugin

MCP server that connects Claude Code to Claude Design. Lets agents send design instructions, read Claude Design replies, pull generated designs into a local repository, and watch Claude Design work in real time. JavaScript. 43 GitHub stars and 6 forks since August 11, 2026.

What it does for you: YY's current UI workflow switches manually between Claude Code for code generation and Claude Design for visual output. claude-design-mcp closes that gap: a single Claude Code session can issue design instructions, wait for Claude Design to produce outputs, and pull the assets back into the AXION repo without leaving the terminal. Relevant for here.now site builds and MCL content pages.

In practice: 43 stars in 6 days for a narrow bridge tool is a genuine niche signal. The watch-it-work capability suggests real-time streaming rather than fire-and-forget polling. If Claude Design access is confirmed, this fills the last manual handoff in the design-to-code pipeline.

For: Claude (Axion). MCP server bridging Claude Code to Claude Design. Claude-specific: requires an active Claude Design subscription alongside Claude Code.

Security4

Quality3

Auditability4

Useful to you3

Useful to community3

Buildable now4

Hermes2

Verdict: watch. 43 stars, precise fit for the Claude Code to Claude Design handoff. Held at WATCH because it requires a Claude Design subscription and YY has not confirmed using Claude Design as a production tool in the current workflow.

Install: npm install and add to Claude Code MCP config per README. Requires Claude Design access. Not currently in YY's settings.json enabledPlugins list. Review before enabling. Copy

Source · 43 GitHub stars and 6 forks as of August 17, 2026. Created August 11, updated August 15. JavaScript implementation. Repository description confirms send-instructions, read-replies, pull-designs, watch-it-work capabilities.

#12

newwatchBoth runtimes26 / 35

Kronop/vibe-aso skill

Claude Code skill for App Store Optimization keyword research for apps created with AI-assisted development. Covers keyword discovery, competitor analysis, and metadata suggestions for iOS and Android. 140 GitHub stars since August 12, 2026.

What it does for you: If any AXION tool, MCL product, or FSX Forum project reaches mobile App Store distribution, vibe-aso provides keyword research and metadata suggestions without a separate ASO subscription. Positioned as the natural follow-on to building an app with AI: once built, the skill handles discoverability.

In practice: 140 stars in 5 days for an ASO skill is notable given the niche audience. The vibe-coding angle is well-timed as more developers ship AI-built apps without an ASO background. Practical and specific.

For: Both runtimes. Claude Code skill, language-agnostic. Works in Claude Code, Codex, and any skill-compatible runner with no code execution required.

Security5

Quality3

Auditability5

Useful to you2

Useful to community3

Buildable now5

Hermes3

Verdict: watch. 140 stars for a useful specialist skill. Held at WATCH because YY has no active App Store release in the current pipeline. Revisit if FSX Forum, RevX, or any MCL tool reaches mobile distribution.

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

Source · 140 GitHub stars as of August 17, 2026. Created August 12, updated August 16. Claude Code skill format confirmed. Repository covers keyword discovery, competitor analysis, and metadata suggestions for App Store and Play Store.

#13

newwatchBoth runtimes25 / 35

zimodzh/dsh-plugin-dev-skills skill

Agent Skills skill covering the standards for developing DeepSeek Harness (DSH) plugins: plugins, services, events, tools, LLM adapters, and packaging and installation patterns. 33 GitHub stars since August 15, 2026. Compatible with Claude Code, Codex, DSH, and VS Code Copilot.

What it does for you: If YY decides to build DSH plugins (the DSH ecosystem appeared in two items this week: dsh-noema and dsh-plugin-dev-skills), this skill teaches the correct DSH plugin architecture standards to the agent writing the code. Without it, an agent building a DSH plugin would infer structure from examples rather than from the authoritative spec.

In practice: 33 stars in 2 days is a modest signal. The real value is as a developer reference that prevents architecture mistakes when building for a new platform. Paired with dsh-noema (item 8), it completes the DSH evaluation picture for YY.

For: Both runtimes. Agent Skills skill compatible with Claude Code, Codex, VS Code Copilot, and DSH. Works in any skill-compatible runner.

Security5

Quality2

Auditability5

Useful to you2

Useful to community3

Buildable now5

Hermes3

Verdict: watch. 33 stars. Held at WATCH because DSH is a new ecosystem for AXION and adopting it is not yet decided. Revisit once YY has evaluated whether DSH fits alongside or replaces the existing Claude Code harness.

Build #13 zimodzh/dsh-plugin-dev-skills: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/zimodzh/dsh-plugin-dev-skills. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · 33 GitHub stars and 1 fork as of August 17, 2026. Created August 15, updated August 16. Topics: agent-skills, awesome-dsh-plugin, claude-code, codex, deepseek-harness, dsh, dsh-plugin. Compatible agent list confirmed in repository description.

#14

watchBoth runtimes24 / 35

pydantic/pydantic-ai repo

PydanticAI agent framework. Version 2.31.0 released August 15, 2026. Changes: UIEventStream can now be built without a run_input, and AGUIEventStream gets its own thread_id and run_id per stream. Bug fix: FallbackModel spans now attribute to the failing model, not the fallback wrapper.

What it does for you: The AGUIEventStream thread_id and run_id change enables independent stream identification in multi-agent pipelines, which is directly relevant to AXION sessions that fan out to sub-agents. The FallbackModel attribution fix makes observability more accurate when AXION uses model fallback chains.

In practice: PydanticAI has shipped 31 minor versions since the June 2026 initial registry entry, confirming active development. v2.31.0 is a focused release with targeted improvements rather than a major architectural change.

For: Both runtimes. Python agent framework; works in any Python environment including AXION bridge and Hermes agent pipelines.

Security4

Quality3

Auditability4

Useful to you3

Useful to community4

Buildable now3

Hermes3

Verdict: watch. Already in registry since June 21, 2026. Re-listed because v2.31.0 (August 15) contains a material change to AGUIEventStream identifiers relevant to multi-agent AXION sessions. No action needed unless actively using AGUIEventStream.

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

Source · Version 2.31.0 released August 15, 2026. Changelog entry confirmed: UIEventStream without run_input, AGUIEventStream own thread_id and run_id, FallbackModel span attribution fix. Originally registered June 21, 2026.

#15

watchBoth runtimes22 / 35

BerriAI/litellm repo

LiteLLM unified LLM proxy and SDK. Version 1.97.0 released August 16, 2026. Key changes: Cursor thinking and fast model-name suffix resolution on the /cursor/chat/completions endpoint, non-streaming response toggle in the playground, team default-organization application to new teams from organization settings, and a fix for team callbacks that previously continued logging after disable_logging was called.

What it does for you: The disable_logging fix is directly relevant to AXION: if YY uses LiteLLM to proxy calls and has disabled logging for a session, the fix ensures logging actually stops rather than silently continuing. The Cursor suffix resolution matters if any AXION tool routes Cursor API calls through the LiteLLM proxy.

In practice: LiteLLM releases at this cadence (version 1.97 in August) confirm it as a continuously maintained production proxy. The fixes in this release are operationally important rather than new features.

For: Both runtimes. Python proxy and SDK layer over 100-plus LLM providers. Works in any Python environment including AXION bridge.

Security3

Quality3

Auditability3

Useful to you3

Useful to community4

Buildable now3

Hermes3

Verdict: watch. Already in registry since June 23, 2026. Re-listed because v1.97.0 (August 16) includes the disable_logging fix that is relevant to AXION logging control. Update if currently on an older version.

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

Source · Version 1.97.0 released August 16, 2026. Changelog confirms: Cursor thinking suffix resolution, non-streaming toggle, team default-organization, disable_logging fix. Originally registered June 23, 2026.

#16

newwatchClaude (Axion)23 / 35

v-yun/vuln-report-skill skill

Claude Code skill that converts confirmed vulnerability findings into submission-ready DOCX reports for bug bounty SRC submissions and 0day disclosures. Includes step-by-step PoC sections, screenshot enforcement rules, and layered verification gates. 23 GitHub stars since August 16, 2026.

What it does for you: When bar181/bar-observatory (item 7) surfaces security anomalies in Claude Code sessions, this skill provides the downstream output format: a submission-ready vulnerability disclosure report. Relevant if AXION surfaces anything in the red-team work that warrants formal reporting.

In practice: 23 stars on a day-old skill targeting a specific professional workflow (bug bounty DOCX reports) is a credible niche signal. The SRC and 0day scoping confirms real-world bug-bounty use rather than a toy example.

For: Claude (Axion). Claude Code skill. Topics confirm claude-code and claude-skill. Produces DOCX output requiring a local DOCX library; less portable to Hermes without that dependency.

Security4

Quality2

Auditability4

Useful to you3

Useful to community3

Buildable now5

Hermes2

Verdict: watch. 23 stars, very new (August 16). Useful if YY does bug bounty or vulnerability disclosure work. Not a current AXION priority; watch for use when security audit outputs require formal reporting.

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

Source · 23 GitHub stars and 0 forks as of August 17, 2026. Created August 16, updated August 16. Topics: bug-bounty, claude-code, claude-skill, docx, security-reporting, src, vulnerability-disclosure. DOCX output format confirmed in description.

#17

newwatchBoth runtimes23 / 35

alphaparkinc/genpark-open-source-voice-agent-pipeline-orchestrator-skill skill

Open-source voice agent pipeline orchestrator skill with streaming ASR and TTS in the Dograh style. Covers pipeline composition for voice input, LLM processing, and voice output. 8 GitHub stars since August 14, 2026.

What it does for you: If YY builds a voice interface for AXION or MCL (for example, a voice-accessible version of the task hub or a voice briefing skill), this skill teaches the pipeline composition pattern for streaming ASR to LLM to TTS without requiring YY to design the architecture from scratch.

In practice: 8 stars is a minimal signal. Included because the voice lens (lens 10) found no better candidates today. The Dograh-style pipeline reference suggests the author has studied production voice agent architectures rather than inventing from scratch.

For: Both runtimes. Claude Code skill covering a voice agent pipeline orchestration pattern. Works in any skill-compatible runner; the actual ASR and TTS integrations require external services.

Security4

Quality2

Auditability4

Useful to you3

Useful to community3

Buildable now4

Hermes3

Verdict: watch. 8 stars. Included to cover the voice lens; no voice agent tool with stronger community signal appeared today. Watch for a better-validated voice pipeline skill before building.

Build #17 alphaparkinc/genpark-open-source-voice-agent-pipeline-orchestrator-skill: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/alphaparkinc/genpark-open-source-voice-agent-pipeline-orchestrator-skill. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · 8 GitHub stars as of August 17, 2026. Created August 14. Topics: voice-agent, streaming, ASR, TTS. Dograh-style pipeline reference confirmed in description.

#18

newwatchBoth runtimes23 / 35

tsuru0805/speak-aloud-mcp plugin

MCP server that lets AI agents produce speech output via ElevenLabs TTS, with volume set and restore controls. macOS and Windows compatible. 10 GitHub stars.

What it does for you: Gives any AXION or Hermes agent the ability to speak its output aloud, which is useful for accessibility, hands-free monitoring of long sessions, or alerting YY when an overnight job completes. Pairs naturally with the heartbeat and notification work already in the bridge.

In practice: 10 stars is a minimal signal but the Windows compatibility and ElevenLabs integration are concrete and useful. Volume set and restore prevents the agent from leaving the system volume in an unexpected state after speaking.

For: Both runtimes. MCP server for ElevenLabs TTS. Any MCP-enabled agent on macOS or Windows can call it to produce speech output during a session.

Security4

Quality2

Auditability4

Useful to you3

Useful to community3

Buildable now4

Hermes3

Verdict: watch. 10 stars. Useful but minimal community validation. Revisit once ElevenLabs TTS use is confirmed for any AXION workflow.

Install: npm install and add MCP server to Claude Code config per README. Requires ElevenLabs API key. Not currently in YY's settings.json enabledPlugins list. Review before enabling. Copy

Source · 10 GitHub stars as of August 17, 2026. macOS and Windows compatibility confirmed in description. ElevenLabs TTS integration and volume set and restore controls confirmed.

#19

newwatchClaude (Axion)19 / 35

jake-j-io2002j1/cupcat-video-editor repo

Free, local, AI-native video editor for Windows where Claude edits the user's video timeline over MCP. 11 GitHub stars since approximately August 13, 2026. Windows-native application.

What it does for you: YY produces AI video for MCL using Higgsfield and Seedance. cupcat-video-editor offers a local Windows video editing interface where Claude Code can directly manipulate the timeline without requiring a paid NLE subscription. Low star count limits confidence but the Windows-native and MCP-driven approach is distinctive.

In practice: 11 stars is a minimal signal. Included because the AI-native video editor concept is directly relevant to MCL's video production workflow and no better Windows-native option appeared today. Watch for community growth before investing setup time.

For: Claude (Axion). Claude edits the video timeline over MCP. Requires Claude as the controlling agent; the editor itself is a Windows-native application.

Security3

Quality2

Auditability3

Useful to you3

Useful to community3

Buildable now3

Hermes2

Verdict: watch. 11 stars, too early to recommend for MCL production use. Watch for growth and external reviews before building into the video pipeline.

Build #19 jake-j-io2002j1/cupcat-video-editor: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/jake-j-io2002j1/cupcat-video-editor. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · 11 GitHub stars as of August 17, 2026. Windows-native confirmed. Claude edits timeline over MCP confirmed in description. Free and local (no subscription) confirmed.

#20

newwatchBoth runtimes24 / 35

evan-steinhilb/md2hd repo

CLI tool that converts Markdown documents into HD (hierarchical diagram) maps for visual comprehension of complex topics and relationships. Designed to be invoked by AI agents to create visual maps from existing documentation. 93 GitHub stars since approximately August 13, 2026. JavaScript.

What it does for you: AXION has extensive Markdown documentation: CLAUDE.md files, brain pages, skill specs, and planning documents. md2hd lets an agent generate visual HD maps from any of these without leaving the terminal, making complex AXION structures easier to reason about in a single session. Relevant for onboarding new agents to AXION's architecture.

In practice: 93 stars in 4 days for a CLI tool with a clear single-purpose design. The agent-invocable framing positions it as infra, not just a user tool, which is the right fit for AXION automation.

For: Both runtimes. JavaScript CLI that any agent can invoke via shell. Runs in any environment with Node.js.

Security4

Quality3

Auditability4

Useful to you3

Useful to community3

Buildable now4

Hermes3

Verdict: watch. 93 stars and useful for AXION documentation visualization. Held at WATCH because it is an enhancement rather than a capability gap filler, and 20 slots are filled with higher-priority items.

Build #20 evan-steinhilb/md2hd: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/evan-steinhilb/md2hd. Save canonical skill/agent under AXION\Skills and AXION\Agents. Copy

Source · 93 GitHub stars as of August 17, 2026. Created approximately August 13, updated August 16. JavaScript CLI. Agent-invocable HD map generation from Markdown confirmed in description.