Chat Skills for AI Agents
One file. Any agent. Working chat in under 10 minutes.
240 followers
One file. Any agent. Working chat in under 10 minutes.
240 followers
Add fully functional chat to your React app using AI agent skills. Drop in a skill file and your agent handles the full integration: installation, environment setup, SSR-safe rendering, and the correct init and login flow. One file per framework. Works seamlessly with Cursor, Kiro, VS Code Copilot, Claude, and Codex. Go from zero to a working chat in minutes, built on CometChat's proven foundation and ready for real projects from day one.
This is the 2nd launch from Chat Skills for AI Agents. View more
Calling Skills for AI Agents
Launched this week
HD voice and video calling by CometChat, built to fit into and grow with your platform. Packed with recording, screen sharing, call logs, raise hand, broadcast mode, picture-in-picture, and more. Integrate via UI Kits, SDKs, or a single npx command (npx @cometchat/skills add) using CometChat Skills. Scales with heavy bandwidth, compliant with global standards, and built for developers.
Launch Team
Chat Skills for AI Agents
Hey Product Hunt! Swapnil here, AVP engineering at CometChat.
A few months ago we launched Chat Skills, a skill file that lets your AI coding agent integrate CometChat's full chat product in under 15 minutes. The response was incredible. Today we're back with the next one: Calling Skills.
v4.2.0 of the CometChat Skills package adds first-class voice and video calling integration across 6 platform families: React, Next.js, React Native, Angular, Android, iOS, and Flutter.
Here's what makes it different from just pointing your agent at the docs:
The dispatcher asks one question before it touches your project: Ringing or Session? These are fundamentally different integration paths. Ringing means a full incoming/outgoing call surface, CallKit on iOS, ConnectionService on Android, VoIP push to wake the device. Session means a link-driven meeting room where both peers join the same session ID, no ringing surface, no Chat SDK dependency. Getting this wrong mid-integration is expensive. The skill resolves it up front.
Once you pick a mode, the agent detects your framework and SDK version, scaffolds the correct file structure, and runs a 23-point verification pass covering VoIP push configuration, SDK initialization order, hangup teardown, permission strings, and API drift issues we caught and fixed across Android, iOS, Flutter, and React Native.
If you're already using Chat Skills, this is fully additive. Same install, same mental model, no changes to your existing chat integration.
Drop your questions below, I'm here all day.
@swapnull Hey Swapnil, congrats on launch number two. The "Ringing or Session, resolve it up front" detail is the part that shows this was built by someone who's actually been burned mid-integration, getting that wrong late is exactly the expensive kind of wrong. The thread you opened that I'd pull harder on is the 23-point pass catching API drift across four platforms. That's the real maintenance nightmare in SDK-first integration: the skill scaffolds correct code against today's SDK versions, but six platform SDKs drift independently and constantly. Three months out, CometChat ships a breaking change or iOS deprecates a permission API, and the integration the agent wrote is now quietly wrong. So is the skill the thing that re-detects and re-scaffolds when an SDK moves, or is it a point-in-time generator and drift becomes the developer's problem again? Because "working chat in 10 minutes" is the easy promise, "still-working chat in 10 months without a rewrite" is the one that decides whether teams trust generated integrations. Following along.
Chat Skills for AI Agents
@artem_fedorovich That’s a very fair callout, and honestly one of the hardest problems in SDK-first integration systems.
Today, each skill is tied to a specific SDK contract - API signatures, permission models, lifecycle requirements, and platform-specific implementation patterns. The 23-point validation pass exists specifically to ensure the generated integration is correct against that contract at generation time.
When the SDK evolves, the skill evolves with it. So rather than being a one-time code generator, the model is closer to a maintained, versioned integration layer. If we introduce a breaking change - for example, a new session lifecycle, updated permissions flow on iOS, or API restructuring - the corresponding skill is updated to reflect the new contract.
From the developer side, the expectation is not “rewrite your integration.” Instead, they can re-run the updated skill against their existing codebase and receive targeted diffs and validation feedback focused only on the affected areas.
Where your observation is absolutely correct is around proactive drift detection.
Today, we do not automatically detect that an existing integration has become stale because of SDK or platform evolution. Developers need to pull the updated skill and re-run validation themselves. That means the system is maintained and evolvable, but not yet fully self-healing.
What we are actively exploring is a compatibility-validation layer that can continuously compare an app’s integration against the latest SDK contract and surface actionable drift warnings proactively - essentially a “your integration is falling behind” signal before things break in production.
So the current philosophy is:
- Not a point-in-time generator
- Not a black-box self-healing system either
- A maintained contract layer with low-cost re-scaffolding and validation
The long-term trust story is exactly what you pointed at: not just “working chat in 10 minutes,” but “still-working chat in 10 months without painful rewrites.” That’s the bar we’re aiming for, and proactive drift detection is likely a key part of getting there.
@swapnull Smart move forcing the Ringing vs Session split upfront. Mixing CallKit/ConnectionService with link-based rooms always causes painful rewrites. That 23-point verification for VoIP push, permissions, and teardown solves the exact native-bridge headaches I’ve debugged across RN and Flutter. Does it auto-handle foreground/background state transitions, or leave it to the host app?
Chat Skills for AI Agents
@nitishkumarpro Foreground and background state management is intentionally left to the host application so developers maintain full control over their app lifecycle and behavior. That said, we already provide implementation guidance and supporting skills for handling these transitions at the application level, helping teams accelerate integration while preserving flexibility.
The SDK-first design here is smart. Wrapping WebRTC into agent-callable tools means the agent owns the session lifecycle rather than punting that complexity to the app layer. At RetainSure we've been building AI workflows for customer success and native call-handling always meant a separate service layer. How are you handling SIP/PSTN bridging for enterprise customers who need traditional telephony alongside IP calling?
Chat Skills for AI Agents
@anand_thakkar1 SIP/PSTN bridging is not supported in the current version of the platform. However, enterprise telephony interoperability is an area we are actively evaluating, and it is part of our broader roadmap as we continue expanding our calling infrastructure for enterprise use cases.
Interesting timing for this launch given how quickly AI agents are moving from chat to real-time interaction.
Curious where you see the biggest adoption curve happening first:
AI sales/support agents handling live customer conversations?
Internal enterprise copilots?
Consumer-facing assistants?
Also wondering how you’re thinking about trust signals in voice/video interactions with AI agents. Do you think the winning platforms will need visible “human handoff” layers and transparency features built directly into the calling experience?
Adding calling via a single npx command is a real DX win. Most teams spend days on integration boilerplate that should be a one-liner. We've been building in the customer success for developer tool companies space, and Calling Skills for AI Agents touches on something we think about a lot. How does the skill handle conflicts with existing auth in apps that already have a communication layer?
Chat Skills for AI Agents
@shivam_jaiswal36 The behavior and adaptability of the skills can vary depending on the underlying model being used. In our testing across multiple models and integration scenarios, we have not encountered conflicts with existing authentication systems or communication layers. The design goal is to work alongside existing app architectures rather than require developers to replace them.
We built something like this from scratch once. The main issue was video latency when participants were far away from each other. You need servers in different parts of the world and a lot of other things. Did you manage to solve this problem somehow?
Chat Skills for AI Agents
@natalia_iankovych otally feel that pain - geo-distributed media is exactly the kind of problem that looks simple on a whiteboard and turns into a full-time ops job in production.
Yes, we handle this. Our calling infrastructure runs on globally distributed media servers. When participants join a call, they're automatically routed through the nearest server node, so you don't get the "one person in Singapore, other in New York, both routing through a single US server" problem.
From the developer's side, there's nothing to configure beyond setting your app's region during SDK init. No TURN server management, no relay logic, no bandwidth tuning. The SDK and our infra handle routing, codec negotiation, and adaptive quality automatically.
The short version: latency between geographically distant participants was a problem we specifically architected around, and it's fully managed infrastructure - not something the integrating team needs to think about or maintain.
The voice-first access model feels especially powerful for support or sales workflows.
When an AI agent detects a query it can't resolve confidently, does it seamlessly transfer to a human with full context, or is there a fallback flow? And can businesses customize the AI's tone/voice to match their brand?
Chat Skills for AI Agents
@diana_nadim2 Today, our focus is on delivering the core calling infrastructure rather than AI-agent-specific voice or video experiences. However, we do believe trust, transparency, and clear escalation paths will become increasingly important as AI-driven communication experiences evolve. Features such as visible human handoff, AI disclosure, and contextual transparency are areas we are closely watching and exploring as part of our long-term roadmap.
@swapnull Love this answer-trust and transparency as core infrastructure, not afterthoughts.
The focus on solid plumbing makes total sense first. When you do explore the agent-facing layer down the line, I'll be watching closely. That balance of capability + clarity is exactly what thoughtful builders need.
Thanks for the thoughtful reply, and congrats again on the launch!
mailX by mailwarm
Does it support real time transcripts and tool events so the agent can act during the call?
Chat Skills for AI Agents
@thamibenjelloun Real-time transcripts and in-call tool event orchestration are not supported today. However, these are capabilities we see as highly valuable for AI-assisted communication workflows, and they are part of our planned roadmap as the platform evolves.