AgentKits

Best toolkits for your AI agents 🤌 🔥

[AgentKits @BestAgentKits]

ClaudeKit is the fastest path from idea to production - built on top of Claude Code - Production-ready AI subagents automate your development & marketing workflows. Build in hours, not weeks.

🤌https://t.co/gIZ3nPCwxf

[AgentKits @BestAgentKits]

ck:llms turns your docs folder into llms.txt: a clean, LLM-readable map of the docs that matter.

less URL guessing, less context waste, better Claude Code runs when agents need project docs before touching code

[AgentKits @BestAgentKits]

copywriting prompts are easy to reroll and hard to trust.

ck:copywriting turns the repeatable parts into rails: formulas, headline patterns, email/landing page structures, CTA checks, and style extraction.

less vibe guessing. more shippable copy loops.

[AgentKits @BestAgentKits]

ck:xlsx is spreadsheet work with rails: formulas over hardcodes, recalc before delivery, zero #REF!/#DIV/0!, and source notes for assumptions

because one broken excel cell can make the whole model cosplay as truth

[AgentKits @BestAgentKits]

ck:pptx turns PowerPoint work into an actual engineering loop

brief -> html layouts -> OOXML edits -> thumbnail validation -> deck

use it when slides need templates, notes, comments, charts, or zero surprise cutoffs

[Duy /zuey/ @goon_nguyen]

Big news: ClaudeKit is now AgentKit.

Same team. Same mission. New name, sharper focus on agent-first developer tools.

Here's everything you need to know 👇

[AgentKits @BestAgentKits]

docx is where "just edit the file" becomes a trap

ck:docx gives Claude Code a real Word workflow: extract text, inspect OOXML, batch redlines, pack the file, then verify the result

useful when legal/business docs need tracked changes, not vibes

[AgentKits @BestAgentKits]

ck:ask is for the moment before you touch code. feed it the architecture question, constraints, tradeoffs, and it forces a systems/strategy/scale/risk pass. useful when the expensive bug is not implementation, but choosing the wrong shape.

[AgentKits @BestAgentKits]

pdf work in claude code should not mean copy-paste archaeology.

ck:pdf gives agents a repeatable loop for extracting text/tables, splitting/merging pages, filling forms, and generating docs without burning context on manual document surgery.

[AgentKits @BestAgentKits]

agentic coding projects do not fail because plan.md exists

they fail because nobody can see the plan while the agent is moving

ck:plans-kanban opens the ClaudeKit /plans dashboard: kanban, timelines, progress, and quick jumps back into plan files

[AgentKits @BestAgentKits]

ck:ai-multimodal turns messy media work into a repeatable Claude Code workflow: analyze images/audio/video/docs with Gemini, generate via Gemini/OpenRouter/MiniMax, and keep provider fallback in scripts instead of vibes.

[AgentKits @BestAgentKits]

ck:gkg gives Claude Code a semantic map of your repo

definitions, usages, impact analysis, repo map

grep finds strings
a code graph shows what might break before the agent refactors it

use it when changing Ruby/Java/Kotlin code with real call-site risk

[AgentKits @BestAgentKits]

ck:docs-seeker gives Claude Code a boring but useful habit: check current docs before coding

it detects the topic, fetches llms.txt docs, then ranks what to read

less hallucinated APIs, fewer stale snippets

[AgentKits @BestAgentKits]

auth is where vibe coding gets expensive.

ck:better-auth turns the messy parts into a repeatable Claude Code loop: database schema, OAuth, sessions, 2FA/passkeys, rate limits, migrations.

less snippet gambling, more gated implementation.

[AgentKits @BestAgentKits]

most agent failures start before code exists

ck:scenario turns a feature into edge cases across 12 dimensions: auth, timing, scale, weird inputs, compliance, integrations, business logic

less vibe-debugging after deploy
more ugly cases caught before the PR

[AgentKits @BestAgentKits]

new ClaudeKit skill: ck:google-adk-python

turn Google ADK into a repeatable agent workbench: root_agent/app structure, MCP tools, A2A, state, evals, then Cloud Run or Vertex deploy

good for builders who want agents that leave the notebook and actually ship

[AgentKits @BestAgentKits]

ck:cook is the skill for when "just code it" would be expensive.

it scouts the repo, pins requirements, writes a reviewed plan, then implements with test/review gates.

use it when the task is clear enough to build, but risky enough to need discipline.

[AgentKits @BestAgentKits]

ck:brainstorm is the skill you run before Claude Code starts touching files.

it forces options, tradeoffs, risks, and an approval gate first.

less "agent, go build the vague thing"
more "show me the plan, the failure modes, then earn the right to code"

[AgentKits @BestAgentKits]

ck:gkg turns code navigation into graph queries

find definitions, usages, and refactor impact before Claude touches files

less random grep, more semantic context

for agents working in Ruby/Java/Kotlin today, with TS/Python still partial

ClaudeKit skill

[AgentKits @BestAgentKits]

ck:google-adk-python turns Google ADK into a Claude Code skill

use it when an agent needs more than a prompt: MCP tools, state/memory, callbacks, evals, and Cloud Run/Vertex deployment paths

prototype less, ship the control plane

[AgentKits @BestAgentKits]

ck:use-mcp is for the moment your agent has 40 MCP tools available and zero reason to load all of them into the main context.

it discovers, catalogs, then executes through Gemini or deterministic scripts.

less tool soup. more controlled runs