suna

How to send a notetaker bot into a video meeting (Google Meet / Zoom / Teams) and work with what it captures — join a call, confirm it got admitted, monitor its lifecycle, pull the speaker-attributed transcript, and produce a recap with decisions + action items, then remove it. Drives the `meet` CLI, which runs through the Kortix Executor (the meeting-bot provider key is resolved server-side; nothing in the sandbox). Load this WHENEVER the user asks to join / attend / record / transcribe / take notes on a meeting, "send a notetaker", "summarize that call", drops a Meet/Zoom/Teams link, or asks how to do anything with a meeting bot.

Kortix brand + design system: the rules, tokens, and component library for building any Kortix frontend UI (apps/web). Load this WHENEVER you create or edit a page, screen, component, list, card, badge, avatar, modal, form, empty state, toast, tooltip, or any visual surface in apps/web. Always load the companion skill make-interfaces-feel-better (apps/web/.agents/skills/make-interfaces-feel-better/SKILL.md) in the same session — brand/tokens here, polish/motion/haptics there. Source of truth: globals.css + the live /design-system page + src/components/ui + the reference implementations listed below.

How to roll Kortix PRODUCTION back to an older already-released version — the inverse of a release. Covers the one-dispatch rollback-prod.yml engine, the per-surface mechanics (API + gateway = Argo image-tag swap; frontend = Vercel promote), the all-important Vercel frontend behavior (why a backend-only push can 'clobber' a FE rollback, and the 'don't rebuild the FE for backend-only pushes' skip that fixes it), the DB/migration-drift safety check that is the real blocker, and how a later promote returns prod to latest. Load WHENEVER the user wants to roll back / revert / downgrade / 'go back a version' on prod, asks how the rollback or the frontend clobber/skip behavior works, or needs to run rollback-prod.yml. Pairs with kortix-release (the forward direction).

Create, manage, validate, preview, and export HTML presentation slides (1920x1080). Load this skill when you need to build a slide deck, export to PDF/PPTX, or preview slides in a browser.

Create, manage, validate, preview, and export HTML presentation slides (1920x1080). Load this skill when you need to build a slide deck, export to PDF/PPTX, or preview slides in a browser.

Canonical reference for a Kortix project: the platform model (repo-native projects, sessions on ephemeral branches, the strict boundary between `kortix.toml` and OpenCode config under `.kortix/opencode/`); the full `kortix.toml` manifest (keys, trigger fields, secrets contract, `[[apps]]` deploy surface); the complete `kortix` CLI (commands, flags, the project-scoped token model, the in-sandbox `KORTIX_TOKEN`); the change-request (CR) system for landing session work on `main` (an agent MUST open a CR to merge); the session sandbox runtime (which supports Docker and Docker-in-Docker); and the OpenCode runtime (agents, skills, commands, tools, plugins, MCP servers, permissions, AGENTS.md rules, models). Load whenever the user asks how Kortix works, about `kortix.toml`, the `kortix` CLI, anything under `.kortix/opencode/`, how to merge/ship/land work on `main`, change requests/CRs/PRs, or to author/edit any OpenCode primitive.

Use for creating, editing, reviewing, and validating presentation decks in PPTX format.

Use for building prospect-specific sales assets such as landing pages, decks, one-pagers, and workflow demos.

Deep research agent skill. Use when the user needs thorough, scientific, truth-seeking research on any topic -- investigating claims, finding primary sources, synthesizing evidence, producing cited reports. Triggers on: 'research this', 'investigate', 'deep dive', 'find sources', 'what does the evidence say', 'literature review', 'fact check', 'analyze the research on', any request requiring multi-source investigation with citations.

Use for structured review, fact-checking, annotation, and correction of documents such as PDF, DOCX, PPTX, and XLSX files.

Use for creating, editing, extracting, converting, and reviewing Word documents.

Use when the user wants a YouTube transcript from a single URL or video ID. Optimized for one input and one output: fetch the transcript fast, default to plain transcript text only, and avoid extra commentary unless the user asks for timestamps, JSON, or metadata. Triggers on: youtube transcript, transcript from this video, get captions, extract transcript from YouTube, summarize this YouTube transcript after fetching it.

Create professional logos through an intelligent, iterative design process. Use this skill when the user wants to create a logo, icon, favicon, brand mark, wordmark, or any visual brand identity mark. Triggers on: 'create a logo', 'design a logo', 'make me a logo', 'logo for my brand', 'I need a logo', 'brand mark', 'wordmark', 'logomark', 'icon design', 'favicon'. This is NOT a one-shot image generator — it researches, strategizes, generates symbols with AI, visually inspects every output, then programmatically composes them with real Google Fonts typography into complete logo systems (logomark, wordmark, combination marks in multiple layouts).

Use for creating, editing, extracting, OCRing, filling, and converting PDF documents.

Use for writing substantial research reports in markdown with inline citations, tables, and optional charts. Best when the user wants a deliverable file plus a concise chat summary.

Use for fullstack web apps built with the bundled Express, Vite, React, Tailwind, and Drizzle template.

Use for distinctive production-grade websites, landing pages, and interactive web experiences with strong design and QA discipline.

Use for spreadsheet creation, analysis, financial models, and polished workbook outputs.

How to do feature work in this repo with git worktrees via `pnpm worktree`. Each worktree gets its own branch, its own block of app ports, its own node_modules, and by default shares the primary checkout's standard local Supabase DB for fast setup; pass `--db` only when a separate Supabase project/data plane is needed. Load WHENEVER starting a feature, bugfix, refactor, experiment, or any change you'll want to run/test in isolation; whenever the user mentions worktrees, isolated/parallel dev instances, running multiple branches at once, or 'spin up a worktree'; and whenever you need the exact non-interactive `pnpm worktree` commands and flags. Enforces: every non-trivial change happens in its own worktree.

Use the Kortix Executor to reach external systems from a session. Prefer the `kortix executor` CLI for agent work, use `@kortix/executor-sdk` for durable TypeScript workflows and reusable skills, and treat the `kortix executor mcp` server as an optional compatibility face. Load whenever the user asks the agent to act in an external app/API, inspect available connectors/tools, add/configure a connector, or work with `[[connectors]]` in kortix.toml.

Canonical reference for a Kortix project: the platform model (repo-native projects, sessions on ephemeral branches, the strict boundary between `kortix.toml` and OpenCode config under `.kortix/opencode/`); the full `kortix.toml` manifest (keys, trigger fields, secrets contract, `[[apps]]` deploy surface); the complete `kortix` CLI (commands, flags, the project-scoped token model, the in-sandbox `KORTIX_TOKEN`); the change-request (CR) system for landing session work on `main` (an agent MUST open a CR to merge); the session sandbox runtime (which supports Docker and Docker-in-Docker); and the OpenCode runtime (agents, skills, commands, tools, plugins, MCP servers, permissions, AGENTS.md rules, models). Load whenever the user asks how Kortix works, about `kortix.toml`, the `kortix` CLI, anything under `.kortix/opencode/`, how to merge/ship/land work on `main`, change requests/CRs/PRs, or to author/edit any OpenCode primitive.

How Kortix end-to-end API tests work and the mandatory test-as-source-of-truth workflow. Load WHENEVER you add, change, or remove an API route, status code, auth gate, request/response shape, or status enum under apps/api/src/** — and whenever you touch anything under tests/ (the ke2e suite) or are asked to add test coverage, run the e2e suite, or understand why CI's coverage gate failed. The suite (tests/) is one clean black-box HTTP suite that runs against a LIVE deployed API with real services; spec/end-to-end.md + the route manifest are the source of truth, enforced by a coverage gate.

How to cut a Kortix production release — the versioning philosophy (when patch vs minor vs major) and the exact flow: derive the release title + notes from the FULL git log since the last release, run the Promote workflow, then deploy + verify prod. Load WHENEVER the user wants to release, promote, cut/ship a version, publish a release, or bump the version. The release notes ARE the public /changelog, so they must be 100% accurate to what shipped.

How to change the database schema in this repo. Current engine: node-pg-migrate, with SQL generated from Drizzle schema changes when possible. Load whenever you add/alter/drop a table, column, enum, index, constraint, RLS/function/grant, or any file under packages/db/migrations.

How to reach a CONNECTED MACHINE (a user's laptop/desktop, or any computer paired over the Agent Computer Tunnel) from a Kortix session — read/write files, run shell commands, and drive the desktop (click/type/screenshot) on that machine. It works through the Executor's `computer` connector (the same connectors/discover/describe/call path as every other integration), so there is no separate tunnel client and no token. Load this when the task is about acting ON a specific physical/remote computer the user has connected ("on my laptop…", "read ~/Downloads on my machine", "run this on my desktop", "click the button on my screen"), or when the user asks how the agent reaches their computer. For files INSIDE this sandbox, just use normal shell/fs — not this.

How to reach third-party systems from a Kortix session via the Executor — one interface to every configured integration (Pipedream, MCP, OpenAPI, GraphQL, HTTP, chat `channel`s like Slack, and connected machines via `computer`), exposed as the `kortix-executor` MCP server's tools (connectors, discover, describe, call). Load whenever the user asks the agent to DO something in an external app/API (send an email, create a Stripe charge, post to Slack, query an internal API, call any SaaS), asks "what integrations/connectors/tools do I have", asks to add/configure a connector, or asks about `[[connectors]]` in kortix.toml. The agent must use the Executor's MCP tools rather than hand-rolling API calls with raw tokens.

The mandatory test-with-every-change discipline for this monorepo. Load WHENEVER you write, change, refactor, or remove ANY code under apps/** or packages/** — a function, class, route, component, hook, schema, migration, config, or bug fix — and whenever you touch anything under tests/, add coverage, run a suite, set up CI, or are asked why a gate failed. Defines which test type each change needs (unit/integration/contract/api/e2e/a11y/visual/perf/security), the exact commands and conventions (co-located bun:test, factories, determinism, no comments), and the CI gates that enforce it. Enforces THE RULE: every possible change ships with tests in the same change.

How to answer in Slack as a teammate. Covers the live plan-block stream (`slack step` with --detail/--output, `slack send` to finalize the answer), file uploads, posting to other channels/threads, reactions, search, message editing/deletion, and the tone the bot should use. Load this when the turn is triggered from Slack (the prompt mentions a Slack workspace/channel/thread, or `$SLACK_CHANNEL_ID` is set in the env), or when the user asks how to do anything in Slack.

Use when creating, editing, extracting, or validating Kortix/Suna .pptx slide decks, pitch decks, or presentations — produces on-brand decks (Kortix colors, Roobert type, symbol logo) and schema-validates them against OOXML before delivery.

Use for fullstack web apps built with the bundled Express, Vite, React, Tailwind, and Drizzle template.

Get a clear view of what matters most today. This skill works with whatever you tell me, and gets richer when you connect your tools.

You are a meeting preparation assistant for an in-house legal team. You gather context from connected sources, prepare structured briefings for meetings with legal relevance, an...

How to read, write, and curate project memory in `.kortix/memory/` — the project brain. Load this skill whenever you (or the memory-reflector agent) need to add, update, or reorganize what this project knows about itself. Defines the rubric for what belongs in memory, the file structure, and the change-request flow for landing memory edits on `main`.

Applies official Kortix colors, type, and layout from kortix.com to decks, docs, slides, and other artifacts. Use when matching brand visuals, formatting deliverables, or checking work against company design standards.

Use when writing or reviewing any Kortix-facing words — headlines, taglines, elevator pitches, audience pitches (developers, companies, enterprise), captions, deck or social copy, product naming, or text composited into images — or when another skill needs Kortix's canonical positioning, terminology, or approved wording. Verbal source of truth; pair with brand-guidelines for visuals.

Generates clean, on-brand Kortix/Suna images — hero, product, social, blog, illustration, and UI assets. Use when the user asks to create, art-direct, or revise images for Kortix, Suna, agents, sandboxes, coding workflows, launch assets, thumbnails, or covers.

Writes platform-native social content — LinkedIn posts, Twitter/X threads, carousels, Instagram/TikTok/Reels/Shorts video scripts, content calendars — and repurposes one long-form piece into many. Use when the user asks for a "LinkedIn post," "Twitter thread," "carousel," "social calendar," "what should I post," "repurpose this," "video hook," "reel," "short-form video," or to grow/strategize a social presence.

Creates and maintains the product marketing context doc at /product-marketing.md — the shared source other marketing skills (copywriting, cro, seo-audit, kortix-image) read first. Use when setting up, auto-drafting, updating, or checking product/marketing context.

Drive Anthropic's Claude Code CLI (`claude -p`) as a non-interactive coding sub-agent from inside Codex. Use when you want to delegate a coding/analysis/refactor task to Claude, get a second opinion / adversarial review from another model, hand off long-context planning, or fan out parallel agents. Covers the exact `claude -p` flags, permission modes, output capture, JSON/stream-json modes, session resume, parallel fan-out, and the mandatory "delegate → capture → independently verify, never trust the self-report" supervision loop.

Drive OpenAI's Codex CLI (`codex exec`) as a non-interactive coding sub-agent from inside Claude Code. Load WHENEVER you want to delegate a coding/analysis/refactor task to Codex, get a second opinion / adversarial review from another model, fan out parallel agents across files or worktrees, or run a long mechanical job while you stay the planner. Covers the exact `codex exec` flags, sandbox tiers, output capture, JSON/schema modes, session resume, parallel fan-out, and the mandatory "delegate → capture → independently verify, never trust the self-report" supervision loop.