Imported from aurokin/agentchat (AGENTS.md). Install upstream with npx skills add aurokin/agentchat. Copyright stays with the author.

Agentchat Agent Instructions

Name: agentchat
Author: aurokin

Read this file before making changes in this repository.

Quick Reference

Package manager: Bun. Use bun install and bun run <script>.
Product name: Agentchat.
Primary surfaces: apps/web, apps/server, apps/mobile, packages/convex, and packages/shared.
Product source of truth: docs/agentchat.
Harness model: Harness Engineering.
Git: avoid git commit --no-verify so hooks run.

Orient with README.md, docs/agentchat/README.md, and Local Modes.
Prepare a fresh checkout with bun install. Populate <repo>/.env.convex.local (apps/server reads it directly via bun --env-file=...; apps/web's next.config.ts loads it at startup). New worktrees pick it up automatically via wt step copy-ignored in the post-start hook.
Use bun dev when a live stack is needed; Ctrl+C to stop. Web + server are routed through portless — portless list or portless get <name> show the current URLs (worktrees get a <branch>. subdomain prefix automatically).
Use wt switch -c <branch> for disposable worktree work; wt remove to tear it down.
Mobile is run separately via bun --cwd apps/mobile dev (Expo is intentionally outside portless until support matures).
Run the health task for each modified surface: bun run health:web, bun run health:server, bun run health:mobile, bun run health:shared, or bun run health:convex.
Use repo gates when touching contracts, env, docs, imports, or shared behavior: bun run env:check, bun run docs:check, bun run lint:repo, bun run verify:ci.

Build for the self-hosted, agent-centric architecture already in progress.
Do not reintroduce hosted-product assumptions, billing, analytics, attachments, or browser-local product data modes.
Treat Convex as the source of truth for auth, conversations, runs, and runtime bindings.
Treat apps/server as the runtime layer. Codex is the active runtime. Pi, OpenCode, and Claude Code are planned.
Runtime is an implementation detail of each agent, not a separate user-facing concept. See Provider-Agent Merge Plan and Roadmap.