foundation: manifesto, origin, governance templates, and Phase 0 planning docs

.github/CODEOWNERS

@@ -0,0 +1,50 @@
+# CODEOWNERS — automatic reviewer assignment
+# https://docs.github.com/en/repositories/managing-your-repositorys-settings-and-security/customizing-your-repository/about-code-owners
+#
+# When a PR touches a matching path, the listed owner(s) are auto-requested for review.
+# Add co-maintainers here as the team grows.
+
+# Default fallback — everything without a more specific rule
+*                                   @EmersonBraun
+
+# Governance and foundation documents
+/MANIFESTO.md                       @EmersonBraun
+/ORIGIN.md                          @EmersonBraun
+/CLAUDE.md                          @EmersonBraun
+/ARCHITECTURE.md                    @EmersonBraun
+/LICENSING.md                       @EmersonBraun
+/docs/architecture/                 @EmersonBraun
+/rfcs/                              @EmersonBraun
+
+# CI, workflows, repo config
+/.github/                           @EmersonBraun
+/turbo.json                         @EmersonBraun
+/pnpm-workspace.yaml                @EmersonBraun
+/.changeset/                        @EmersonBraun
+
+# Core package — the sacred one
+/packages/core/                     @EmersonBraun
+
+# Per-package ownership (extend as contributors take responsibility)
+/packages/adapters/                 @EmersonBraun
+/packages/react/                    @EmersonBraun
+/packages/ink/                      @EmersonBraun
+/packages/cli/                      @EmersonBraun
+/packages/runtime/                  @EmersonBraun
+/packages/tools/                    @EmersonBraun
+/packages/skills/                   @EmersonBraun
+/packages/memory/                   @EmersonBraun
+/packages/rag/                      @EmersonBraun
+/packages/sandbox/                  @EmersonBraun
+/packages/observability/            @EmersonBraun
+/packages/eval/                     @EmersonBraun
+/packages/templates/                @EmersonBraun
+
+# Documentation site
+/apps/docs/                         @EmersonBraun
+
+# Example apps
+/apps/example-react/                @EmersonBraun
+/apps/example-ink/                  @EmersonBraun
+/apps/example-runtime/              @EmersonBraun
+/apps/example-multi-agent/          @EmersonBraun

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,8 @@
+blank_issues_enabled: false
+contact_links:
+  - name: Question or help
+    url: https://github.com/EmersonBraun/agentskit/discussions
+    about: For general questions, use GitHub Discussions.
+  - name: Security disclosure
+    url: https://github.com/EmersonBraun/agentskit/blob/main/SECURITY.md
+    about: For security issues, follow the private disclosure process.

.github/ISSUE_TEMPLATE/docs.yml

@@ -0,0 +1,36 @@
+name: Documentation
+description: Report a docs problem or request new documentation
+labels: ["type-docs"]
+body:
+  - type: dropdown
+    id: kind
+    attributes:
+      label: What kind of docs issue?
+      options:
+        - Incorrect content
+        - Missing content
+        - Unclear explanation
+        - Broken example
+        - Broken link
+        - Translation issue
+    validations:
+      required: true
+  - type: input
+    id: page
+    attributes:
+      label: Page URL or path
+      placeholder: "https://docs.agentskit.io/concepts/adapter or apps/docs/docs/..."
+    validations:
+      required: true
+  - type: textarea
+    id: description
+    attributes:
+      label: What is the problem?
+      description: Describe what's wrong, missing, or unclear.
+    validations:
+      required: true
+  - type: textarea
+    id: suggestion
+    attributes:
+      label: Suggested fix (optional)
+      description: If you have an idea for how to fix it, share it here.

.github/ISSUE_TEMPLATE/rfc.yml

@@ -0,0 +1,65 @@
+name: RFC — Request for Comments
+description: Propose a significant change, new contract, or architectural decision
+labels: ["rfc", "discussion"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        RFCs are for non-trivial changes: new contracts, new packages, breaking changes, architectural shifts.
+        For small features or bug fixes, use the **Feature Request** or **Bug Report** templates.
+
+        See the [RFC process](../blob/main/rfcs/README.md) before opening.
+  - type: textarea
+    id: summary
+    attributes:
+      label: Summary
+      description: One paragraph explanation of the proposal.
+    validations:
+      required: true
+  - type: textarea
+    id: motivation
+    attributes:
+      label: Motivation
+      description: Why are we doing this? What problem does it solve? Who benefits?
+    validations:
+      required: true
+  - type: textarea
+    id: proposal
+    attributes:
+      label: Detailed proposal
+      description: The core of the RFC. Include contracts, APIs, examples, pseudocode.
+    validations:
+      required: true
+  - type: textarea
+    id: alternatives
+    attributes:
+      label: Alternatives considered
+      description: What other approaches were evaluated? Why not them?
+    validations:
+      required: true
+  - type: textarea
+    id: drawbacks
+    attributes:
+      label: Drawbacks and unresolved questions
+      description: Be honest about cost, complexity, migration burden, open questions.
+    validations:
+      required: true
+  - type: dropdown
+    id: impact
+    attributes:
+      label: Impact
+      options:
+        - Breaking change (major bump)
+        - Additive (minor bump)
+        - Internal (no user impact)
+    validations:
+      required: true
+  - type: checkboxes
+    id: checks
+    attributes:
+      label: Pre-submission checklist
+      options:
+        - label: I have read the [Manifesto](../blob/main/MANIFESTO.md) and this proposal aligns with it
+          required: true
+        - label: I have searched existing RFCs and issues for similar proposals
+          required: true

.github/PULL_REQUEST_TEMPLATE.md

@@ -1,18 +1,39 @@
+<!-- Thanks for contributing! Please fill out this template. -->
+
 ## What
 
-Brief description of the change.
+<!-- Brief description of the change. -->
 
 ## Why
 
-What problem does this solve? Link to issue if applicable.
+<!-- What problem does this solve? Link to issue: Closes #NNN -->
 
 ## How
 
-How was this implemented? Any important design decisions?
+<!-- Implementation notes. Non-trivial design decisions, trade-offs, alternatives considered. -->
+
+## Type of change
+
+- [ ] Bug fix (non-breaking)
+- [ ] New feature (non-breaking)
+- [ ] Breaking change (requires RFC reference below)
+- [ ] Docs only
+- [ ] Chore / internal
+
+## RFC / ADR reference
+
+<!-- If this is a breaking change or a new contract, link the RFC or ADR. -->
 
 ## Checklist
 
-- [ ] Tests pass (`npm test`)
-- [ ] Types check (`npm run lint`)
-- [ ] Docs updated (if public API changed)
-- [ ] Follows existing code style
+- [ ] Tests added or updated (`pnpm test` passes)
+- [ ] Types check (`pnpm lint`)
+- [ ] Bundle size within budget (`pnpm size` — when configured)
+- [ ] Changeset created (`pnpm changeset`) — for any user-facing change
+- [ ] Docs updated (concept page, recipe, or API reference)
+- [ ] Manifesto principles respected (core <10KB, plug-and-play, zero lock-in, no `any`)
+- [ ] Screenshots or demo recorded (for UI changes)
+
+## Screenshots / demo
+
+<!-- Optional but appreciated for UI or DX changes. -->

MANIFESTO.md

@@ -0,0 +1,68 @@
+# AgentsKit Manifesto
+
+> These are the non-negotiable principles that shape AgentsKit. Every design decision, every PR, every package is measured against them. If a principle is ever about to be broken, we stop and reconsider — never rationalize.
+
+---
+
+## 1. The core is sacred
+
+`@agentskit/core` will always be **under 10KB gzipped with zero runtime dependencies**. It contains only types, contracts, and minimal orchestration primitives. Nothing gets added here unless it is foundational and permanent. Stability is the feature.
+
+## 2. Every package is plug-and-play
+
+Every package works standalone, with a single install and less than ten lines of configuration to produce something useful. If a package requires another package to function at all, we have an architecture problem — not a dependency problem.
+
+## 3. Interop is radical, not optional
+
+Any combination of AgentsKit packages must compose without glue code. React + Runtime + Memory + RAG + Observability should just work. The contract is the API — shared types, shared events, shared semantics across the entire ecosystem.
+
+## 4. Zero lock-in
+
+Leaving AgentsKit must be a single `npm uninstall` away. No proprietary formats, no captive state, no hidden coupling. Adapters, memory stores, tools — all are replaceable. We earn our place every day.
+
+## 5. Agent-first, not chat-first
+
+AgentsKit is not a chat library that happens to support tools. It is an agent runtime that happens to render chat. Tools, skills, memory, and reasoning loops are primary citizens — UI is one surface among many (React, Ink, CLI, headless).
+
+## 6. Docs are product, not afterthought
+
+Documentation is not written after shipping. Every feature lands with: a concept page, a recipe, an API reference, and a troubleshooting entry. If it isn't documented, it doesn't exist.
+
+## 7. TypeScript rigor, no escape hatches
+
+Strict mode always. `any` is forbidden — use `unknown` and narrow. Public APIs have inferred, provable types. Schema-first means `Tool` signatures *become* `.execute` return types without a second declaration.
+
+## 8. Small, deep, testable modules
+
+Every module is deep: a simple, stable interface hiding meaningful behavior. Tests target external contracts, not implementation details. When a module shows up in multiple test files as a mock, it should become a real primitive.
+
+## 9. Predictable is better than clever
+
+A predictable API that takes ten minutes to learn beats a clever one that takes ten hours. If a new contributor can't understand a module in one sitting, we refactor — not rename.
+
+## 10. Open by default
+
+Roadmap is public. RFCs are public. ADRs are public. Post-mortems are public. Decisions made in private die in silence.
+
+---
+
+## What we will never do
+
+- Ship a package that imports `@agentskit/core` and inflates it
+- Break a contract without a deprecation cycle and a migration guide
+- Add a feature that works only on one provider
+- Write documentation in PR descriptions and hope users find it
+- Accept "it works on my machine" — E2E or it didn't happen
+- Invent a concept where a standard already exists (prefer MCP, A2A, JSON Schema, OpenAPI)
+
+## What we will always do
+
+- Read the manifesto before saying "yes" to a new feature
+- Prefer deleting code to adding code
+- Deprecate loudly, deprecate early, deprecate with a path forward
+- Treat every release as a promise to someone building a business on top
+
+---
+
+*Last updated: 2026-04-14.*
+*Changes to this manifesto require an RFC and maintainer consensus.*

ORIGIN.md

@@ -0,0 +1,57 @@
+# Origin — Why AgentsKit Exists
+
+> *Personal note from the author. This document is intentionally first-person and unpolished. It is the story of a frustration that became a library.*
+>
+> *To the maintainer reading this before merge: replace the placeholders in square brackets with your real experience, or rewrite freely. The goal is authenticity — not a perfect pitch.*
+
+---
+
+## The frustration
+
+In april, 2026, I was trying to build a simple thing: an AI chat interface with streaming, a couple of tools, and persistent memory. Not an agent that cures cancer — just a chat that remembers yesterday's conversation and can call a function.
+
+I went shopping in the JavaScript ecosystem.
+
+**Vercel AI SDK** got me 80% of the way in an hour. Streaming — excellent. Tool calling — clean. Then I wanted an autonomous loop, multi-agent delegation, a runtime that survives a crash. Nothing. Vercel AI SDK is a beautiful *chat SDK*. It is not an *agent framework*, and it never pretended to be.
+
+**LangChain.js** had all the abstractions I wanted. It also had 200MB of dependencies, leaked abstractions at every layer, and every example on the internet was outdated within six weeks. I spent two days debugging a `RunnableSequence` and writing no product.
+
+**assistant-ui** had beautiful components. I counted 53 before I gave up trying to figure out which one I needed.
+
+**Mastra** was promising and new — exactly my problem. I loved the direction but the pieces I wanted most weren't there yet.
+
+**MCP** solved tool interop brilliantly. It solved nothing else. You still had to write the UI, the runtime, the memory, the streaming.
+
+So my weekend project became: *paste LangChain code into Claude, remove half of it, add bits of Vercel AI SDK, wire up a homebrew ReAct loop, glue a Pinecone client onto the side.* Three days in, I had something that worked. Four days in, I couldn't debug it. Five days in, I deleted it all.
+
+## The observation
+
+The JavaScript ecosystem has given us a beautiful pattern, over and over again: a small, opinionated core with a radically composable ecosystem around it. React. Express. Vite. Fastify. Drizzle. Hono.
+
+The agent space has not yet found its React. It has a monolith (LangChain), a narrow product (Vercel AI SDK), component libraries without runtimes, runtimes without UI, and a protocol (MCP) with no opinion about the application layer.
+
+The observation is simple: **we don't need another framework. We need a kit** — a family of small, well-contracted packages where every piece works alone and every combination works together.
+
+## The bet
+
+AgentsKit is a bet on three claims:
+
+1. **JavaScript will be the language of agent applications**, because it's where the users are, where the deployment is cheap, and where the ecosystem is most plug-and-play.
+2. **Agents don't need a framework; they need contracts.** A `Tool` is a function. A `Memory` is a store. A `Runtime` is a loop. Formalize the contracts, keep the packages small, let users assemble.
+3. **Plug-and-play beats opinionated all-in-one.** If you outgrow any package, swap it. If you don't need one, don't install it. The core is 10KB and stays that way.
+
+## The promise
+
+AgentsKit will never become the thing I was frustrated with. It will stay small. It will stay composable. It will stay honest — about what it does and what it doesn't. It will tell you when you should use something else.
+
+If AgentsKit ever starts feeling like LangChain — bloated, magical, hard to debug — we failed. If it ever stops feeling like plain JavaScript you already know — we failed. The manifesto is how we hold ourselves accountable.
+
+## The invitation
+
+If you've felt the frustration above, this is for you. If you've built agents in JavaScript and thought "there has to be a better way" — this is that way, or at least our attempt at it.
+
+Open an issue. Open an RFC. Build a package. Break our assumptions. Make us better.
+
+This is day one.
+
+— Emerson Braun, 04 april 2026