Hunexia Skill and Plugin Stack (Adoption Recommendation)

As of 2026-07-15. A curation of the sprawling global skill ecosystem, fit to Hunexia's real practice. Core premise: Most skills are already installed globally. The real problem is not "installation" but curation, authentication, and official adoption. Per the brand tone (no exaggeration; show the evidence), we clearly separate "right now" from "conditional."

---

Current Composition (Survey Result)

Section	Status
Active plugins (2)	pr-review-toolkit · chrome-devtools-mcp (auth0 removed, 8329cc9)
Custom	5 skills (/hire-agent, /org-review, /generate-proposal, /content-calendar, /automation-audit), 13 agents (data-analyst hired)
.mcp.json	None. Project-level MCP is only via the chrome-devtools plugin
§7 referenced MCP real composition	Notion ✅ authenticated (2026-07-15). Google Workspace (Gmail, Calendar, Drive) on hold (Google Cloud OAuth server setup needed). Supabase, Vercel, Figma decided unused
hooks	None

Capability gaps (closure status): ① Data analysis / reporting ✅ data-analyst hired ② Doyun / Yuna web access ✅ granted ③ MCP — Notion ✅, Google Workspace on hold.

---

Tier 1 — Adopt Immediately (High Value, Low Cost)

Items already in the global install or referenced by §7. The essence is official adoption (listed in §7) + authentication, not "installation."

#	Item	Detail	Effect	Pros	Cons / Caution
1	MCP authentication and installation	Notion ✅ authenticated (2026-07-15). Supabase, Vercel, Figma = decided unused. Gmail, Calendar, Drive = on hold (Google Cloud OAuth server setup)	ops-pm documents, automation-rag RAG actually run	Notion runs immediately with one-time auth	Google MCP requires a Google Cloud OAuth client (cost ongoing)
2	deep-research	Multi-source web search → fact-check → cited report	Higher trust in diagnosis, sales evidence, and content. Directly serves the brand tone "no exaggeration; show the evidence"	Used across ai-adoption, sales, content	Token / time cost is high → only for big questions
3	dataviz	Consistently generate charts/dashboards with a brand-neutral palette	Higher quality in diagnosis tables, proposals, and content visuals. Fills part of the "data analysis gap"	Already referenced by automation-audit and generate-proposal	Palette swap to Hunexia tokens (navy / mint / violet) needed
4	make-pdf + document-generate	Markdown → publication-quality PDF / generate missing documents automatically	Enables PDF delivery of customer proposals and course materials	Already referenced by generate-proposal and course-content	Increases dependency on the gstack family
5	commit-commands	Commit / push / PR, prune gone branches	Unifies the workflow that records every organizational change as git HR history	Standardizes people-ops commits	Almost none
6	fewer-permission-prompts	Analyzes usage patterns → suggests an allowlist for read-only calls	Reduces repeated permission prompts	Current settings.local.json allow has only 1 entry → large effect	Initial one-time scan required
7	skill-creator	Skill creation / modification / evaluation / optimization	Maintains the 5 custom skills; skill evolution on /org-review	Aligns with the roadmap (agentskills.io standard migration)	None
8	claude-md-management	Audit and improve CLAUDE.md	Maintains hub-document health (§7 registry updates, etc.)	Automated	Almost none

---

Tier 2 — Conditional Adoption (On Trigger)

Item	Trigger	Detail	Pros	Cons / Caution
vercel suite (deploy, AI SDK, next, etc.)	Launching our own web product	Deployment, preview, rollback, AI app architecture	Core when Labs launches a web service (business area)	Currently only the design-system package → on hold
supabase skill	RAG / backend in earnest	Postgres best practices, schema	Aligns with automation-rag's pgvector work	MCP auth already done in Tier 1
huggingface-skills	Labs model research	Model selection, training, Spaces	Supports RAG / model evaluation	Very heavy → select only what's needed
chrome-devtools-mcp (just installed)	Labs web QA	Real-browser debugging, performance / LCP	Verifies web-service quality	Overlaps playwright MCP → recommend unifying to one
runway-api	YouTube video production in earnest	Video / image / audio generation API	content-strategist video assets	Cost + brand tone (avoid exaggeration) → careful
hookify	When procedural automation is needed	Converts rules to hooks	Automates brand-tone checks and approval gates via hooks	Misconfiguration has side effects → use with careful / guard

---

Tier 3 — Just-installed Plugin Evaluation (Decision Needed)

Plugin	Evaluation	Basis
pr-review-toolkit	✅ Keep (Labs)	Reinforces the Labs chain with PR review (/review-pr). However, overlaps with feature-dev:code-reviewer, code-review, and gstack:review → recommend unifying to one team standard
chrome-devtools-mcp	✅ Keep (Labs web QA)	Debugging / performance measurement. However, playwright MCP is already global → split roles (playwright = automation / Chrome DT = debug) or unify
auth0	⚠️ Suitability in question → recommend removal	Unnecessary unless we directly launch an authentication SaaS. Current output = design-system package. If Phase 4 "Hunexia Agents" productization is not imminent, remove (reduces clutter in the 60+ skill list)

Code-review duplication unification plan: At least 4 exist globally — code-review (top-level) · code-review:code-review (pr-review-toolkit) · gstack:review · feature-dev:code-reviewer (agent). Pick one Labs standard to prevent confusion. Standard: feature-dev:code-reviewer agent + code-review skill. (This standard is reflected in CLAUDE.md §7.)

---

Tier 4 — Recommended Unused (Skip / Removal Candidates)

• save-to-spotify: Unrelated to Hunexia.
• gstack ios-*: No iOS products.
• Many remaining education-agent-skills: Academy agents (curriculum-architect, course-content-writer) already use a curated subset → no additions needed.

---

Strategic Decisions (Human Judgment Needed)

1. Whether to remove auth0 (Tier 3): If no launch plan, remove.
2. New hire for data analysis / reporting agent: A gap the organization itself identified. dataviz + deep-research fill the tools, but there is no responsible agent (e.g., data-analyst) → recommend /hire-agent.
3. brand-content-designer / course-content-writer web access: Currently no WebSearch / WebFetch. Decide via /org-review whether to add tools (allow independent research) or maintain division of labor through content-strategist.

---

Execution Phases

1. /reload-plugins — 2 active plugins (pr-review-toolkit · chrome-devtools-mcp; auth0 removed in 8329cc9).
2. MCP — Notion ✅ authenticated (2026-07-15). Supabase, Vercel, Figma = decided unused.
3. Google Workspace MCP (Gmail, Calendar, Drive) on hold — needs Google Cloud OAuth server setup (cost ongoing).
4. CLAUDE.md §7 "reuse" registry lists Tier 1 skills officially — ✅ done (2026-07-15).
5. Remove auth0 → ✅ done (removed from .claude/settings.json, 8329cc9).
6. Unify code-review standard to 1 — ✅ done (stated in §7).
7. Capability gaps → ✅ done: hired data-analyst (/hire-agent), granted Doyun / Yuna web access (/org-review).
8. Save this document (knowledge/org/skills-stack.md) durably — ✅ done.

Verification

• No effect on design-system code: Skills / plugins do not touch source code → no change to npm run typecheck.
• Plugin load: After /reload-plugins, 3 are shown.
• MCP smoke test: Notion page read returned normally ✅ (2026-07-15). (Google Workspace is on hold.)
• §7 update: New skills and standard review tool listed in the registry — ✅.

---

Per-role Skill Mapping (Reflected in org-review 2026-07-15 — Coaching)

Skills stated in each agent's Process. [Core] = already usable (autonomously invoked via Skill / Agent tool addition). (Conditional) = delegate after /plugin install. Academy skills are selected from the frentis-surveyed set (education-agent-skills-adoption) in education-agent-skills — pipeline structure not adopted, only skill knowledge.

Role	[Core] Autonomously Invoked Skills	(Conditional · after install)
orchestrator	(Routing-only — has Agent; minimal skills)	—
people-ops	skill-creator · claude-md-management · remember/learn (Skill); agent-creator (Agent)	—
content-strategist	deep-research · dataviz · make-pdf · frontend-design · /content-calendar	exa/tavily (research) · postiz (SNS) · save-to-spotify (podcast) · runway-api (video)
brand-content-designer	frontend-design · ui-ux-pro-max · figma · playground · dataviz · make-pdf · design-system	canva · adobe-for-creativity · miro · runway-api
sales-proposal-expert	deep-research · dataviz · make-pdf · /generate-proposal	hunter/apollo/lusha (leads) · monday-crm/intercom (CRM)
ai-adoption-consultant	deep-research · dataviz · /automation-audit	zapier (automation mapping) · amplitude/windsor-ai (analytics)
curriculum-architect	Education design: backwards-design-unit-planner · scope-and-sequence-designer · curriculum-knowledge-architecture-designer · competency-unpacker · kud-chart-author · learning-target-authoring-guide · learning-progression-builder · criterion-referenced-rubric-generator · assessment-validity-checker · hinge-question-designer · formative-assessment-technique-selector · differentiation-adapter · udl-barrier-anticipator/udl-options-designer · project-brief-designer · coverage-audit · prompt-literacy-sequence-designer (AI course) ＋deep-research · make-pdf	—
course-content-writer	Education content: explicit-instruction-sequence-builder · three-part-lesson-designer · lesson-opening-designer · think-aloud-script-generator · digital-worked-example-sequence · worked-example-fading-designer · practice-problem-sequence-designer · self-explanation-prompt-designer · cognitive-load-analyser · dual-coding-designer · retrieval-practice-generator · spaced-practice-scheduler · elaborative-interrogation-generator · feedback-quality-analyser · metacognitive-prompt-library · error-analysis-protocol · socratic-questioning-sequence-generator · ai-hallucination-fact-check-protocol (AI course) · pedagogical-content-knowledge-developer (TPACK) ＋frontend-design · design-system · make-pdf	—
principal-architect	feature-dev (architect/explorer) · gstack:spec · diagram · superpowers (writing-plans/brainstorming)	greptile/sourcegraph/serena (code search)
implementation-lead	superpowers (test-driven-development/subagent-driven-development) · feature-dev:code-reviewer · code-review · simplify · verify · commit-commands · security-review	typescript-lsp (TS design system)
automation-rag-engineer	supabase · feature-dev (Agent) · deep-research	huggingface-skills · pinecone/qdrant (vector DB) · firecrawl/exa (web collection) · zapier · mcp-server-dev
ops-pm	(Read/Write locked — depends on MCP: notion / Calendar / Gmail)	linear/asana (task tracking)

Note: context7 is an MCP server (web document retrieval), not a Skill tool target — for up-to-date library / API docs, check the official docs via WebFetch (per actual agent behavior, reflected in e470cb1). All other skills in the table are Skill tool calls.

Remaining Gaps (Handled Separately)

• Security: ✅ security-review added to implementation-lead.
• Data analysis / reporting owner: ❌ No agent → handle separately via /hire-agent (data-analyst).
• Minor (on hold): orchestrator → dispatching-parallel-agents, course-writer → instructional-coaching/panel-review, ops-pm → document-generate. Revisit at a future /org-review.