diff --git a/.agentic/ONRAMP.md b/.agentic/ONRAMP.md
index 920a310..aceb5a3 100644
--- a/.agentic/ONRAMP.md
+++ b/.agentic/ONRAMP.md
@@ -8,13 +8,13 @@ This repo dogfoods its own toolbelt. When you work on `@tnezdev/agentic`, you us
 
 ```bash
 # 1. Orient — load the maintainer hat with live situational tokens
-bun src/cli/main.ts persona activate spores-maintainer
+bun packages/agentic/src/cli/main.ts persona activate spores-maintainer
 
 # 2. Pick up work — top ready task from the local queue
-bun src/cli/main.ts task next
+bun packages/agentic/src/cli/main.ts task next
 
 # 3. When cutting a release — run the checklist
-bun src/cli/main.ts skill run release-check
+bun packages/agentic/src/cli/main.ts skill run release-check
 ```
 
 Read the persona output. Do what step 3 of its "Before picking up work" section tells you. Start the work.
diff --git a/.agentic/README.md b/.agentic/README.md
index e3822b2..e0f8b8f 100644
--- a/.agentic/README.md
+++ b/.agentic/README.md
@@ -22,28 +22,28 @@ From the repo root:
 
 ```bash
 # List what's available
-bun src/cli/main.ts persona list
-bun src/cli/main.ts skill list
-bun src/cli/main.ts workflow list
-bun src/cli/main.ts task list
+bun packages/agentic/src/cli/main.ts persona list
+bun packages/agentic/src/cli/main.ts skill list
+bun packages/agentic/src/cli/main.ts workflow list
+bun packages/agentic/src/cli/main.ts task list
 
 # Activate the maintainer hat — pipe into your LLM of choice
-bun src/cli/main.ts persona activate spores-maintainer
+bun packages/agentic/src/cli/main.ts persona activate spores-maintainer
 
 # Run the release check skill
-bun src/cli/main.ts skill run release-check
+bun packages/agentic/src/cli/main.ts skill run release-check
 
 # Pick up the next ready task
-bun src/cli/main.ts task next
+bun packages/agentic/src/cli/main.ts task next
 
 # Kick off a release run
-bun src/cli/main.ts workflow run agentic-release --name "0.5.0-cut"
+bun packages/agentic/src/cli/main.ts workflow run agentic-release --name "0.5.0-cut"
 
 # Query memories when you need the "why"
-bun src/cli/main.ts memory recall "runtime scope"
+bun packages/agentic/src/cli/main.ts memory recall "runtime scope"
 ```
 
-(After `npm install -g @tnezdev/agentic`, replace `bun src/cli/main.ts` with `agentic`.)
+(After `npm install -g @tnezdev/agentic`, replace `bun packages/agentic/src/cli/main.ts` with `agentic`.)
 
 ## Why this shape
 
diff --git a/.agentic/personas/spores-maintainer.md b/.agentic/personas/spores-maintainer.md
index f88118b..e8dc8a2 100644
--- a/.agentic/personas/spores-maintainer.md
+++ b/.agentic/personas/spores-maintainer.md
@@ -20,11 +20,11 @@ The time is `{{timestamp}}`.
 
 ## Principles (non-negotiable)
 
-- **Zero production dependencies.** `"dependencies": {}` in `package.json` stays clean. devDependencies only when absolutely required.
-- **Types first.** Add shapes to `src/types.ts` before writing implementations.
+- **Zero production dependencies.** `"dependencies": {}` in `packages/agentic/package.json` stays clean. devDependencies only when absolutely required.
+- **Types first.** Add shapes to `packages/agentic/src/types.ts` before writing implementations.
 - **Adapter pattern.** Every primitive has an `adapter.ts` interface and a `filesystem.ts` implementation. New storage backends implement the same interface.
-- **No `console.log` in library code.** CLI output goes through `output(ctx, data, formatter)` re-exported from `src/cli/main.ts`.
-- **Test files colocated.** `src/foo/filesystem.test.ts` next to `src/foo/filesystem.ts`. Inline fixtures, no separate fixtures directory.
+- **No `console.log` in library code.** CLI output goes through `output(ctx, data, formatter)` re-exported from `packages/agentic/src/cli/main.ts`.
+- **Test files colocated.** `packages/agentic/src/foo/filesystem.test.ts` next to `packages/agentic/src/foo/filesystem.ts`. Inline fixtures, no separate fixtures directory.
 - **Identity lives outside Agentic.** Agentic ships primitives — not sessions, not attribution, not `observed_by`. If you're tempted to add an identity field, stop.
 
 ## Before picking up work
diff --git a/.agentic/skills/release-check/skill.md b/.agentic/skills/release-check/skill.md
index d4e7e98..1429cd5 100644
--- a/.agentic/skills/release-check/skill.md
+++ b/.agentic/skills/release-check/skill.md
@@ -18,9 +18,9 @@ Do this before the release PR or tag, not while a publish run is failing. The pa
 
 Open a `chore: release vX.Y.Z` PR that:
 
-- Bumps `version` in `package.json` per semver.
+- Bumps `version` in `packages/agentic/package.json` per semver.
 - Appends a CHANGELOG entry (or release notes section) covering user-visible changes since the last tag.
-- Keeps `"dependencies": {}` in `package.json`. A non-empty production dependency is a design regression — investigate before shipping.
+- Keeps `"dependencies": {}` in `packages/agentic/package.json`. A non-empty production dependency is a design regression — investigate before shipping.
 
 Merge only after CI (`.github/workflows/ci.yml`) is green on the PR. CI runs `bun test` and `bun run typecheck` — those are the authoritative gates, not a local run.
 
diff --git a/.agentic/skills/smoke-test/skill.md b/.agentic/skills/smoke-test/skill.md
index 3a96c8a..64da9e4 100644
--- a/.agentic/skills/smoke-test/skill.md
+++ b/.agentic/skills/smoke-test/skill.md
@@ -17,15 +17,15 @@ bash scripts/smoke-test-node.sh
 
 ## What it does
 
-1. `npm pack` in the repo root — produces the exact tarball that `npm publish` would upload
+1. `npm pack` in `packages/agentic` — produces the exact tarball that `npm publish` would upload
 2. Creates a temp consumer project and installs the tarball via `bun add`
 3. Runs `scripts/smoke-consumer.mjs` which imports the public API and checks that all value exports are present and constructable
 
 ## When to run
 
 - Before pushing a release tag (`vX.Y.Z`)
-- After changing `package.json` fields (`files`, `main`, `exports`)
-- After adding or removing public exports from `src/index.ts`
+- After changing `packages/agentic/package.json` fields (`files`, `main`, `exports`)
+- After adding or removing public exports from `packages/agentic/src/index.ts`
 
 CI runs these automatically in `.github/workflows/ci.yml` and `.github/workflows/publish.yml` between the test step and the publish step.
 
diff --git a/.github/workflows/publish.yml b/.github/workflows/publish.yml
index 70db653..9bffa2e 100644
--- a/.github/workflows/publish.yml
+++ b/.github/workflows/publish.yml
@@ -56,4 +56,5 @@ jobs:
           npm --version
 
       - name: Publish
+        working-directory: packages/agentic
         run: npm publish --provenance --access public
diff --git a/.spores/ONRAMP.md b/.spores/ONRAMP.md
index 679443c..b68b5a0 100644
--- a/.spores/ONRAMP.md
+++ b/.spores/ONRAMP.md
@@ -8,13 +8,13 @@ This repo dogfoods its own toolbelt. When you work on `@tnezdev/agentic`, you us
 
 ```bash
 # 1. Orient — load the maintainer hat with live situational tokens
-bun src/cli/main.ts persona activate spores-maintainer
+bun packages/agentic/src/cli/main.ts persona activate spores-maintainer
 
 # 2. Pick up work — top ready task from the local queue
-bun src/cli/main.ts task next
+bun packages/agentic/src/cli/main.ts task next
 
 # 3. When cutting a release — run the checklist
-bun src/cli/main.ts skill run release-check
+bun packages/agentic/src/cli/main.ts skill run release-check
 ```
 
 Read the persona output. Do what step 3 of its "Before picking up work" section tells you. Start the work.
diff --git a/.spores/README.md b/.spores/README.md
index 476da68..64a790d 100644
--- a/.spores/README.md
+++ b/.spores/README.md
@@ -22,28 +22,28 @@ From the repo root:
 
 ```bash
 # List what's available
-bun src/cli/main.ts persona list
-bun src/cli/main.ts skill list
-bun src/cli/main.ts workflow list
-bun src/cli/main.ts task list
+bun packages/agentic/src/cli/main.ts persona list
+bun packages/agentic/src/cli/main.ts skill list
+bun packages/agentic/src/cli/main.ts workflow list
+bun packages/agentic/src/cli/main.ts task list
 
 # Activate the maintainer hat — pipe into your LLM of choice
-bun src/cli/main.ts persona activate spores-maintainer
+bun packages/agentic/src/cli/main.ts persona activate spores-maintainer
 
 # Run the release check skill
-bun src/cli/main.ts skill run release-check
+bun packages/agentic/src/cli/main.ts skill run release-check
 
 # Pick up the next ready task
-bun src/cli/main.ts task next
+bun packages/agentic/src/cli/main.ts task next
 
 # Kick off a release run
-bun src/cli/main.ts workflow run spores-release --name "0.1.0-cut"
+bun packages/agentic/src/cli/main.ts workflow run spores-release --name "0.1.0-cut"
 
 # Query memories when you need the "why"
-bun src/cli/main.ts memory recall "runtime scope"
+bun packages/agentic/src/cli/main.ts memory recall "runtime scope"
 ```
 
-(After `npm install -g @tnezdev/agentic`, replace `bun src/cli/main.ts` with `agentic`; `spores` remains a compatibility alias.)
+(After `npm install -g @tnezdev/agentic`, replace `bun packages/agentic/src/cli/main.ts` with `agentic`; `spores` remains a compatibility alias.)
 
 ## Why this shape
 
diff --git a/.spores/personas/spores-maintainer.md b/.spores/personas/spores-maintainer.md
index b059001..7df6fc9 100644
--- a/.spores/personas/spores-maintainer.md
+++ b/.spores/personas/spores-maintainer.md
@@ -20,11 +20,11 @@ The time is `{{timestamp}}`.
 
 ## Principles (non-negotiable)
 
-- **Zero production dependencies.** `"dependencies": {}` in `package.json` stays clean. devDependencies only when absolutely required.
-- **Types first.** Add shapes to `src/types.ts` before writing implementations.
+- **Zero production dependencies.** `"dependencies": {}` in `packages/agentic/package.json` stays clean. devDependencies only when absolutely required.
+- **Types first.** Add shapes to `packages/agentic/src/types.ts` before writing implementations.
 - **Adapter pattern.** Every primitive has an `adapter.ts` interface and a `filesystem.ts` implementation. New storage backends implement the same interface.
-- **No `console.log` in library code.** CLI output goes through `output(ctx, data, formatter)` re-exported from `src/cli/main.ts`.
-- **Test files colocated.** `src/foo/filesystem.test.ts` next to `src/foo/filesystem.ts`. Inline fixtures, no separate fixtures directory.
+- **No `console.log` in library code.** CLI output goes through `output(ctx, data, formatter)` re-exported from `packages/agentic/src/cli/main.ts`.
+- **Test files colocated.** `packages/agentic/src/foo/filesystem.test.ts` next to `packages/agentic/src/foo/filesystem.ts`. Inline fixtures, no separate fixtures directory.
 - **Identity lives outside spores.** Spores ships primitives — not sessions, not attribution, not `observed_by`. If you're tempted to add an identity field, stop.
 
 ## Before picking up work
diff --git a/.spores/skills/release-check/skill.md b/.spores/skills/release-check/skill.md
index 8757688..7ac30c5 100644
--- a/.spores/skills/release-check/skill.md
+++ b/.spores/skills/release-check/skill.md
@@ -16,9 +16,9 @@ Releases are **CI-gated**. You don't run `npm publish`; you push a `vX.Y.Z` tag
 
 Open a `chore: release vX.Y.Z` PR that:
 
-- Bumps `version` in `package.json` per semver.
+- Bumps `version` in `packages/agentic/package.json` per semver.
 - Appends a CHANGELOG entry (or release notes section) covering user-visible changes since the last tag.
-- Keeps `"dependencies": {}` in `package.json`. A non-empty production dependency is a design regression — investigate before shipping.
+- Keeps `"dependencies": {}` in `packages/agentic/package.json`. A non-empty production dependency is a design regression — investigate before shipping.
 
 Merge only after CI (`.github/workflows/ci.yml`) is green on the PR. CI runs `bun test` and `bun run typecheck` — those are the authoritative gates, not a local run.
 
diff --git a/.spores/skills/smoke-test/skill.md b/.spores/skills/smoke-test/skill.md
index 7f4ccab..185c56a 100644
--- a/.spores/skills/smoke-test/skill.md
+++ b/.spores/skills/smoke-test/skill.md
@@ -16,15 +16,15 @@ bash scripts/smoke-test.sh
 
 ## What it does
 
-1. `npm pack` in the repo root — produces the exact tarball that `npm publish` would upload
+1. `npm pack` in `packages/agentic` — produces the exact tarball that `npm publish` would upload
 2. Creates a temp consumer project and installs the tarball via `bun add`
 3. Runs `scripts/smoke-consumer.ts` which imports the public API and checks that all value exports are present and constructable
 
 ## When to run
 
 - Before pushing a release tag (`vX.Y.Z`)
-- After changing `package.json` fields (`files`, `main`, `exports`)
-- After adding or removing public exports from `src/index.ts`
+- After changing `packages/agentic/package.json` fields (`files`, `main`, `exports`)
+- After adding or removing public exports from `packages/agentic/src/index.ts`
 
 CI runs this automatically in `.github/workflows/publish.yml` between the test step and the publish step.
 
diff --git a/AGENTS.md b/AGENTS.md
index 8e3cf65..5d350a4 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -1,252 +1,28 @@
-# AGENTS.md — Agentic (formerly SPORES)
+# AGENTS.md — Agentic Monorepo
 
-Orientation for agent sessions. Concise. Read before touching code.
+Orientation for agent sessions. Read this before touching code in this repo.
 
-> **Rename in progress:** The package is migrating from `spores` / `.spores/` to `agentic` / `.agentic/`. Both names work during the compatibility window. Prefer `agentic` in new code and docs.
+## Start Here
 
-## Start here
+This repo dogfoods its own toolbelt. Before touching code, run the three-command on-ramp in [`.agentic/ONRAMP.md`](.agentic/ONRAMP.md).
 
-This repo dogfoods its own toolbelt. Before touching code, run the three-command on-ramp in [`.agentic/ONRAMP.md`](.agentic/ONRAMP.md) — it activates the `spores-maintainer` persona, pulls the top ready task, and points you at the release skill. The rest of this file is reference; ONRAMP.md is the path.
+## Layout
 
-## What is SPORES?
-
-A TypeScript library + CLI for agent in-loop primitives:
-
-1. **Memory** — remember/recall/dream with L1/L2/L3 tiers
-2. **Skills** — load and run skill.md files from `.agentic/skills/`
-3. **Workflow** — digraph runtime (GraphDef → Run → Transitions, state derived from history)
-4. **Tasks** — typed adapter interface (ULID IDs, Taskwarrior-shaped)
-5. **Persona** — activate a hat at the start of a turn: metadata (memory_tags, skills, task_filter, workflow, routing hints) + a rendered body with live situational facts. Declarative attention, not enforced scope.
-6. **Artifact** — versioned content blobs with metadata (type, title, tags, finalized). The canonical place to store an agent's durable outputs within a project — briefs, memos, reports, plans. See "Artifact primitive" below.
-7. **Source** — pluggable read-only loader abstraction (`read(name) → text`, `list() → names`). Personas/skills/workflows all load through the same shape. `LayeredSource` composes for seed-then-emerge (e.g. live DB shadows seed filesystem). See "Source abstraction" below.
-8. **Dispatch** — foundation types for the universal inbound message primitive (`Dispatch`, `DispatchFilter`, `matchDispatch`). Spores ships the message shape and pure match logic; runtimes ship transport, scheduling, and handler execution.
-
-MVP scope = what an agent reaches for *inside a single turn*. No hosting, no webhooks, no session layer — those are daemon-level concerns. **Identity lives outside spores** — in the run orchestration layer. Spores provides the hat; the caller provides who's wearing it.
-
-## Tech stack
-
-- TypeScript on Bun. No build step. `bun run <file>` directly.
-- Zero production dependencies (`"dependencies": {}` in package.json)
-- Hand-rolled TOML parser in `config.ts`
-- Custom arg parser in `src/cli/main.ts` — no CLI framework
+| Path | Purpose |
+|---|---|
+| `packages/agentic` | Published `@tnezdev/agentic` core package |
+| `.agentic/` | Repo-local dogfood workspace |
+| `docs/` | Repo-level design docs |
+| `scripts/` | Release and smoke-test helpers |
 
 ## Commands
 
-```bash
-bun test          # run all tests
-bun run typecheck # tsc --noEmit
-```
-
-## Architecture
-
-### Types first
-
-All shared types live in `src/types.ts`. Add types there before writing implementations.
-
-### Adapter pattern
-
-Every primitive has an interface in `src/<module>/adapter.ts`. Filesystem implementations are the default. Future storage backends implement the same interface.
-
-| Module | Interface | Implementation |
-|--------|-----------|----------------|
-| memory | implicit in filesystem.ts | `src/memory/filesystem.ts` |
-| workflow | `WorkflowAdapter` | `src/workflow/filesystem.ts` |
-| tasks | `TaskAdapter` | `src/tasks/adapter.ts` (stub only) |
-| personas | `PersonaAdapter` | `src/personas/filesystem.ts` |
-| artifacts | `ArtifactAdapter` | `src/artifact/filesystem.ts` |
-
-### Source abstraction
-
-Config-style primitives (personas, skills, workflows, file-style dispatch configs) load through a pluggable `Source` interface in `src/sources/`:
-
-```typescript
-interface Source {
-  read(name: string): Promise<SourceRecord | undefined>  // text + locator
-  list(): Promise<string[]>
-}
-```
-
-Reference implementations:
-
-- `FlatFileSource(dir, ext)` — `<dir>/<name><ext>` layouts (personas, workflows)
-- `NestedFileSource(dir, filename)` — `<dir>/<name>/<filename>` layouts (skills)
-- `InMemorySource(records, tag)` — for tests and bake-in seed templates
-- `LayeredSource([liveSource, seedSource])` — first-wins read, union-dedupe list (the seed-then-emerge primitive)
-
-Per-primitive load functions (`loadPersonaFromSource`, `loadSkillFromSource`, `loadGraphFromSource`) accept any `Source` — Compass and other remote runtimes plug in their own (DB, HTTP) without touching spores.
-
-Data primitives (Memory, Artifacts, Tasks) have query semantics and live behind their own adapter shapes — `Source` is for config, not data.
-
-### CLI: two-word dispatch
-
-```
-agentic <noun> <verb> [args]    # preferred
-spores  <noun> <verb> [args]    # compatibility alias
-```
-
-Commands in `src/cli/commands/<noun>.ts`. Each command is a `Command` function exported as `<noun><Verb>Command`. The dispatch table is in `src/cli/main.ts`.
-
-Current command surface:
-- `agentic init` — scaffold `.agentic/` config
-- `agentic memory remember/recall/forget/dream/reinforce`
-- `agentic skill list/show/run`
-- `agentic workflow list/show/run/status`
-- `agentic persona list/view/activate`
-- `agentic artifact create/read/write/edit/inspect/list/finalize`
-
-### Skills on disk
-
-```
-~/.agentic/skills/<name>/skill.md    # global (user-level)
-.agentic/skills/<name>/skill.md      # project-level (wins on name conflict)
-# Legacy: ~/.spores/ and .spores/ are honoured when .agentic/ is absent
-```
-
-Frontmatter: `name`, `description`, `tags: [a, b, c]`
-Body: the skill content returned by `skill run` (pipe to an LLM).
-
-### Personas on disk
-
-```
-~/.agentic/personas/<name>.md        # global (user-level)
-.agentic/personas/<name>.md          # project-level (wins on name conflict)
-# Legacy: ~/.spores/ and .spores/ are honoured when .agentic/ is absent
-```
-
-Flat-file layout (unlike skills which use a directory per skill). Frontmatter: `name`, `description`, `memory_tags: [...]`, `skills: [...]`, optional `task_filter: { tags: [...], status: ready }` (nested, one level deep), optional `workflow: <graph-id>`. Body is markdown with `{{cwd}}`, `{{timestamp}}`, `{{hostname}}`, `{{git_branch}}` tokens that get substituted at `persona activate` time.
-
-**`view` vs `activate`** is load-bearing: `view` prints the raw file with literal tokens (for humans editing or reviewing); `activate` substitutes live situational facts (for piping into an LLM). Don't let them produce identical output.
-
-**One hat at a time.** Personas don't compose, stack, or inherit. To pivot, deactivate one and activate another. Runtime integration for applying persona bindings (using `memory_tags` as a recall filter, etc.) is **the caller's responsibility** — spores ships the metadata, the caller wires it. Descoped from v0.1 intentionally; expected to land after we have more signal from actual use.
-
-### Config resolution (four-tier)
-
-1. Hardcoded defaults in `config.ts`
-2. `~/.agentic/config.toml` — global user overrides (preferred); falls back to `~/.spores/config.toml`
-3. `.agentic/config.toml` — project overrides, wins over global (preferred); falls back to `.spores/config.toml`
-
-### Workflow runtime
-
-- `GraphDef` defines a digraph (nodes + edges with conditions)
-- `expandGraph` flattens nested subgraphs at register time — nesting is free
-- `Runtime` is a **state machine only** — it derives current state from `Run.history` (no `current_node` field). It does NOT schedule or evaluate `EvaluatorRef` conditions — that's the caller's job.
-- State is immutable: each transition appends to `history`
-
-### Dispatch
-
-Foundation only — the message shape and pure match logic. `Dispatch` carries `from`, `to`, `payload`, `timestamp`, plus optional `when` / `recurrence` delivery metadata. `DispatchFilter` is a declarative predicate over `from` / `to`; `matchDispatch(dispatch, filter)` returns boolean. `DispatchHandlerHooks` types `onRegister` / `onUnregister` for handler-level lifecycle setup (idempotency lives in the hook, not in spores).
-
-Runtimes own send/handle/cancel verbs, the actual transport, scheduling, and handler execution. The split — vocabulary in spores, execution in runtime — keeps spores callable from both long-running daemons and serverless deployments.
-
-See `PROJECTS/spores/DESIGN-runtime-description.md` for the full design conversation.
-
-### URI schemes
-
-`agentic://` is the preferred URI scheme for Agentic-owned compute. Branded type `AgenticUri = \`agentic://\${string}\`` in `types.ts`. Referenced from skill bodies, dispatched by the host runtime.
-
-`spores://` is the legacy scheme; `SporesUri` is preserved but deprecated. Use `AgenticUri` in new code.
-
-## Conventions
-
-- Test files colocated: `src/memory/filesystem.test.ts` next to `src/memory/filesystem.ts`
-- Test fixtures: inline (no separate fixtures dir)
-- IDs: ULIDs via monotonic factory (see tasks types)
-- Error handling: functions throw on unexpected errors; return `undefined` for "not found" cases (e.g. `loadSkill` returns `undefined` when skill doesn't exist)
-- No `console.log` in library code — CLI output goes through `output(ctx, data, formatter)` in `src/cli/main.ts`
-- **Descriptions are agent-facing activation triggers, not labels.** For both skills and personas, phrase `description` as "Activate when..." rather than "The X maintainer". `list` output is meant to function as a lookup table an agent scans to decide what to reach for — good triggers make the scan useful.
-
-### Artifact primitive
-
-Artifacts are versioned markdown blobs — the agent's durable output store for a project.
-
-**On-disk layout** (inside `.agentic/artifacts/`):
-
-```
-.agentic/artifacts/<ulid>/meta.json    — ArtifactRecord (type, title, version, finalized, tags, …)
-.agentic/artifacts/<ulid>/v1.md        — body at version 1
-.agentic/artifacts/<ulid>/v2.md        — body at version 2 (iterate write)
-```
-
-`body_ref` in `meta.json` is relative to `.agentic/artifacts/` — e.g. `"<id>/v2.md"`.
-Legacy: `.spores/artifacts/` is used when `.agentic/artifacts/` does not exist.
-
-**Write modes:**
-
-| Mode | Behavior |
-|------|----------|
-| `iterate` (default) | Bump version, write `v<n+1>.md`. Prior versions remain on disk. |
-| `replace` | Overwrite current `v<n>.md` in place. Version unchanged. |
-| `create` | Fail with "already exists". Used only for first-write semantics. |
-
-**Finalize semantics:** `artifact finalize <id>` sets `finalized=true` in `meta.json`. Finalized artifacts reject `write` and `edit`. `finalize` is idempotent — finalizing an already-finalized artifact is a no-op.
-
-**CLI worked example:**
+Run checks from the repository root:
 
 ```bash
-# Create
-agentic artifact create brief "## Q2 Launch\n\nTBD." --title "Q2 Brief" --tags "q2,launch"
-
-# Read (pipe-friendly — raw body to stdout in human mode, JSON with --json)
-agentic artifact read 01JXYZ... | pbcopy
-
-# Iterate
-agentic artifact write 01JXYZ... "## Q2 Launch\n\nUpdated content." --mode iterate
-
-# Targeted edit
-agentic artifact edit 01JXYZ... --old "TBD." --new "Final copy."
-
-# Inspect metadata
-agentic artifact inspect 01JXYZ... --json
-
-# List with filter
-agentic artifact list --type brief --json
-
-# Finalize the artifact
-agentic artifact finalize 01JXYZ...
+bun test
+bun run typecheck
+bun run build
 ```
 
-**Hook events:**
-
-| Event | Fired by |
-|-------|---------|
-| `artifact.created` | `artifact create` |
-| `artifact.written` | `artifact write`, `artifact edit` |
-| `artifact.finalized` | `artifact finalize` |
-
-Hook env vars: `AGENTIC_ARTIFACT_ID`, `AGENTIC_ARTIFACT_TYPE`, `AGENTIC_ARTIFACT_TITLE`, `AGENTIC_ARTIFACT_TAGS` (create); `AGENTIC_ARTIFACT_ID`, `AGENTIC_ARTIFACT_VERSION`, `AGENTIC_ARTIFACT_MODE` (written); `AGENTIC_ARTIFACT_ID`, `AGENTIC_ARTIFACT_FINAL_VERSION` (finalized). Legacy `SPORES_*` mirrors are set alongside for compatibility.
-
-**Dogfood hook:** `.agentic/hooks/artifact.written` — indexes the artifact reference into memory after every write so it's searchable via `agentic memory recall`.
-
-**Workflow → artifact worked example:**
-
-A workflow node declares its output type via `artifact.type` (from #50):
-
-```yaml
-nodes:
-  - id: write-brief
-    label: Write Q2 Brief
-    artifact:
-      type: brief
-      description: Q2 launch brief produced by the briefing workflow
-```
-
-After the workflow run completes and the node produces its output, the caller persists it:
-
-```bash
-# Node output lands in the transition; caller writes it as a named artifact
-BODY=$(agentic workflow status "$RUN_ID" --json | jq -r '.nodes[-1].artifact.content')
-ARTIFACT_ID=$(agentic artifact create brief "$BODY" --title "Q2 Brief" --tags "briefing,q2" --json | jq -r '.artifact.id')
-
-# artifact.written hook fires automatically → reference indexed into memory
-# Later retrieval
-agentic memory recall "Q2 Brief"
-```
-
-The dogfood hook wires the last step without any special-cased plumbing in spores itself.
-
-## What NOT to add
-
-- Sessions, webhooks, hosting — daemon-layer, not SPORES
-- Any adapter implementation for tasks until the interface is settled
-- Dependencies — keep `"dependencies": {}` clean
-- Per-module adapter interfaces for memory (memory follows the filesystem.ts shape directly, no separate adapter.ts)
+The root package is private and only coordinates workspaces. Core package implementation rules live in [`packages/agentic/AGENTS.md`](packages/agentic/AGENTS.md).
diff --git a/README.md b/README.md
index a538614..39a9afe 100644
--- a/README.md
+++ b/README.md
@@ -1,92 +1,23 @@
-# @tnezdev/agentic
+# Agentic Monorepo
 
-> **Compatibility window.** This package was formerly `@tnezdev/spores`. New code should use `@tnezdev/agentic`, `agentic`, `.agentic/`, and `AGENTIC_*`. The `spores` CLI alias, `.spores/`, and `SPORES_*` remain supported during migration.
+This repository contains the Agentic package workspace.
 
-**In-loop primitives for agents that run more than once.**
+## Packages
 
-Most agent tooling optimizes for the single prompt. Agentic optimizes for the agent that wakes up tomorrow — with memory, skills, workflows, tasks, and a persona that shape what it does next. It's the toolbelt inside the turn, not the runtime around it.
+| Package | Path | Purpose |
+|---|---|---|
+| `@tnezdev/agentic` | `packages/agentic` | Core primitives and CLI |
 
-Agentic is a primitive layer, not a host runtime. It gives your agent building blocks — store and recall memories, load skills, advance workflows, track tasks, activate personas — without owning scheduling, auth, transport, or execution. That's your daemon's job. Agentic just makes it easier to build.
+Runtime packages will live beside the core package under `packages/`.
 
-Zero production dependencies. Runs on Bun. Works from the CLI.
+## Development
 
-## Why this exists
-
-Agents that run in production need more than a prompt and a model call. They need:
-
-- **Memory** that persists across sessions — not just a growing context window.
-- **Skills** that are loaded on demand, not baked into a system prompt.
-- **Workflows** that define multi-step processes with real state machines, not ad-hoc prompt chains.
-- **Tasks** that track commitments inside the loop, not just on an external board.
-- **Personas** that activate a coherent hat — memory tags, skill set, task filter, and rendered instructions — in one declaration.
-
-Agentic ships all five as filesystem-backed primitives your agent can reach for inside any turn. No hosting, no webhooks, no session layer required.
-
-## Install
+Run checks from the repository root:
 
 ```bash
-npm install @tnezdev/agentic
-# or: bun add @tnezdev/agentic
-```
-
-## Quick start
-
-```bash
-# Scaffold .agentic/ directory in your project
-agentic init
-
-# Store a memory
-agentic memory remember "always emit types from the public API" --tags style,api
-
-# Load a skill
-agentic skill list
-agentic skill run release-check | llm
-
-# Track a task
-agentic task add "update CHANGELOG before tagging"
-agentic task next
-
-# Activate a persona (pipe into your LLM as system prompt)
-agentic persona list
-agentic persona activate spores-maintainer | llm --system -
-```
-
-## Primitives
-
-| Primitive | What it does |
-|-----------|-------------|
-| **Memory** | Tiered store (L1/L2/L3) with recall, reinforce, and dream consolidation |
-| **Skills** | Load `.md` skill files and pipe their content into an LLM |
-| **Workflow** | Directed-graph runtime — register a graph, create runs, advance node state |
-| **Tasks** | ULID-keyed task queue with status transitions and annotations |
-| **Persona** | Activate a "hat" at the start of a turn: memory tags, skills, task filter, workflow, and a rendered body with live situational facts |
-
-All five primitives read from `.agentic/` in your project root, with optional global overrides from `~/.agentic/`. Legacy `.spores/` and `~/.spores/` paths are honoured during the compatibility window when `.agentic/` is absent.
-
-## What Agentic is not
-
-Agentic is intentionally scoped. It does not ship:
-
-- Hosted sessions or transcript storage
-- Webhook receivers or background schedulers
-- Provider clients or credential management
-- Approval inboxes or auth flows
-- Platform-specific task adapters in core
-
-Those are host-runtime responsibilities. Agentic provides the vocabulary and primitives that make hosts easier to build, test, and reason about. See [`docs/framework-boundaries.md`](docs/framework-boundaries.md) for the full boundary design.
-
-## Flags
-
-```
---json          Output as JSON (machine-readable, no truncation)
---wide          Disable column truncation in list output
---base-dir      Override working directory
+bun test
+bun run typecheck
+bun run build
 ```
 
-## Working example
-
-The `.agentic/` directory in this repo is the dogfood — a persona, a skill, a workflow, tasks, and memories that are used to build this package itself. Read [`.agentic/ONRAMP.md`](.agentic/ONRAMP.md) for a tour.
-
-## Architecture
-
-See [AGENTS.md](AGENTS.md) for the full architecture, on-disk layout, and conventions for agent sessions.
+The dogfood workspace still lives at `.agentic/`. Start with [`.agentic/ONRAMP.md`](.agentic/ONRAMP.md), then read [`packages/agentic/AGENTS.md`](packages/agentic/AGENTS.md) for core package architecture.
diff --git a/bun.lock b/bun.lock
index d0aac9c..dd2fbc9 100644
--- a/bun.lock
+++ b/bun.lock
@@ -3,7 +3,15 @@
   "configVersion": 0,
   "workspaces": {
     "": {
+      "name": "agentic-workspace",
+    },
+    "packages/agentic": {
       "name": "@tnezdev/agentic",
+      "version": "0.5.0",
+      "bin": {
+        "agentic": "dist/cli/main.js",
+        "spores": "dist/cli/main.js",
+      },
       "devDependencies": {
         "@cloudflare/workers-types": "^4.20260426.1",
         "@types/bun": "^1.3.11",
@@ -14,6 +22,8 @@
   "packages": {
     "@cloudflare/workers-types": ["@cloudflare/workers-types@4.20260426.1", "", {}, "sha512-cBYeQaWwv/jFV8ualmwp6wIxmAf0rDe2DPPQwPbslKmPHqgv861YpAvm45r05K40QboZgxNQVIPgNkmtHqZeJQ=="],
 
+    "@tnezdev/agentic": ["@tnezdev/agentic@workspace:packages/agentic"],
+
     "@types/bun": ["@types/bun@1.3.11", "", { "dependencies": { "bun-types": "1.3.11" } }, "sha512-5vPne5QvtpjGpsGYXiFyycfpDF2ECyPcTSsFBMa0fraoxiQyMJ3SmuQIGhzPg2WJuWxVBoxWJ2kClYTcw/4fAg=="],
 
     "@types/node": ["@types/node@25.5.2", "", { "dependencies": { "undici-types": "~7.18.0" } }, "sha512-tO4ZIRKNC+MDWV4qKVZe3Ql/woTnmHDr5JD8UI5hn2pwBrHEwOEMZK7WlNb5RKB6EoJ02gwmQS9OrjuFnZYdpg=="],
diff --git a/docs/capability-contracts.md b/docs/capability-contracts.md
index dad18bb..3e07815 100644
--- a/docs/capability-contracts.md
+++ b/docs/capability-contracts.md
@@ -344,7 +344,7 @@ Those are host bindings. The portable layer defines the shared vocabulary that l
 
 ## Neutral Example Set
 
-The following examples are useful fixtures for validating the vocabulary without binding it to a specific host. Each is stored as a JSON file under `src/capability/fixtures/` and validated in tests using `validateCapability`.
+The following examples are useful fixtures for validating the vocabulary without binding it to a specific host. Each is stored as a JSON file under `packages/agentic/src/capability/fixtures/` and validated in tests using `validateCapability`.
 
 ### `issue_tracker.list_issues`
 
diff --git a/docs/lifecycle-events.md b/docs/lifecycle-events.md
index e7c6cbe..fbd8116 100644
--- a/docs/lifecycle-events.md
+++ b/docs/lifecycle-events.md
@@ -2,7 +2,7 @@
 
 Lifecycle events are Agentic-owned vocabulary for semantic facts about primitive operations. They answer "what happened?" without deciding who handles it, where it is stored, or how side effects run.
 
-Agentic defines the event names and portable envelope in `src/types.ts`:
+Agentic defines the event names and portable envelope in `packages/agentic/src/types.ts`:
 
 - `LifecycleEventName`
 - `LifecycleEvent`
diff --git a/package.json b/package.json
index 4bffece..7af2192 100644
--- a/package.json
+++ b/package.json
@@ -1,45 +1,13 @@
 {
-  "name": "@tnezdev/agentic",
-  "version": "0.5.0",
-  "description": "Executable toolbelt for agent self-improvement",
-  "license": "MIT",
+  "name": "agentic-workspace",
+  "private": true,
   "type": "module",
-  "repository": {
-    "type": "git",
-    "url": "git+https://github.com/tnezdev/agentic.git"
-  },
-  "author": "Travis Nesland",
-  "publishConfig": {
-    "access": "public"
-  },
-  "main": "./dist/index.js",
-  "types": "./dist/index.d.ts",
-  "exports": {
-    ".": {
-      "types": "./dist/index.d.ts",
-      "default": "./dist/index.js"
-    }
-  },
-  "bin": {
-    "agentic": "dist/cli/main.js",
-    "spores": "dist/cli/main.js"
-  },
-  "files": [
-    "dist/**/*.js",
-    "dist/**/*.d.ts",
-    "dist/**/*.map",
-    "AGENTS.md",
-    "CHANGELOG.md"
+  "workspaces": [
+    "packages/*"
   ],
   "scripts": {
-    "build": "tsc -p tsconfig.build.json && chmod 755 dist/cli/main.js",
-    "typecheck": "tsc --noEmit",
-    "test": "bun test",
-    "prepack": "bun run build"
-  },
-  "devDependencies": {
-    "@cloudflare/workers-types": "^4.20260426.1",
-    "@types/bun": "^1.3.11",
-    "typescript": "^5.7.0"
+    "build": "bun run --cwd packages/agentic build",
+    "typecheck": "bun run --cwd packages/agentic typecheck",
+    "test": "bun run --cwd packages/agentic test"
   }
 }
diff --git a/.npmignore b/packages/agentic/.npmignore
similarity index 100%
rename from .npmignore
rename to packages/agentic/.npmignore
diff --git a/packages/agentic/AGENTS.md b/packages/agentic/AGENTS.md
new file mode 100644
index 0000000..ddbd79a
--- /dev/null
+++ b/packages/agentic/AGENTS.md
@@ -0,0 +1,252 @@
+# AGENTS.md — Agentic (formerly SPORES)
+
+Orientation for agent sessions. Concise. Read before touching code.
+
+> **Rename in progress:** The package is migrating from `spores` / `.spores/` to `agentic` / `.agentic/`. Both names work during the compatibility window. Prefer `agentic` in new code and docs.
+
+## Start here
+
+This repo dogfoods its own toolbelt. Before touching code, run the three-command on-ramp in [`../../.agentic/ONRAMP.md`](../../.agentic/ONRAMP.md) — it activates the `spores-maintainer` persona, pulls the top ready task, and points you at the release skill. The rest of this file is reference; ONRAMP.md is the path.
+
+## What is SPORES?
+
+A TypeScript library + CLI for agent in-loop primitives:
+
+1. **Memory** — remember/recall/dream with L1/L2/L3 tiers
+2. **Skills** — load and run skill.md files from `.agentic/skills/`
+3. **Workflow** — digraph runtime (GraphDef → Run → Transitions, state derived from history)
+4. **Tasks** — typed adapter interface (ULID IDs, Taskwarrior-shaped)
+5. **Persona** — activate a hat at the start of a turn: metadata (memory_tags, skills, task_filter, workflow, routing hints) + a rendered body with live situational facts. Declarative attention, not enforced scope.
+6. **Artifact** — versioned content blobs with metadata (type, title, tags, finalized). The canonical place to store an agent's durable outputs within a project — briefs, memos, reports, plans. See "Artifact primitive" below.
+7. **Source** — pluggable read-only loader abstraction (`read(name) → text`, `list() → names`). Personas/skills/workflows all load through the same shape. `LayeredSource` composes for seed-then-emerge (e.g. live DB shadows seed filesystem). See "Source abstraction" below.
+8. **Dispatch** — foundation types for the universal inbound message primitive (`Dispatch`, `DispatchFilter`, `matchDispatch`). Spores ships the message shape and pure match logic; runtimes ship transport, scheduling, and handler execution.
+
+MVP scope = what an agent reaches for *inside a single turn*. No hosting, no webhooks, no session layer — those are daemon-level concerns. **Identity lives outside spores** — in the run orchestration layer. Spores provides the hat; the caller provides who's wearing it.
+
+## Tech stack
+
+- TypeScript on Bun. No build step. `bun run <file>` directly.
+- Zero production dependencies (`"dependencies": {}` in `packages/agentic/package.json` stays clean)
+- Hand-rolled TOML parser in `config.ts`
+- Custom arg parser in `src/cli/main.ts` — no CLI framework
+
+## Commands
+
+```bash
+bun test          # run all tests
+bun run typecheck # tsc --noEmit
+```
+
+## Architecture
+
+### Types first
+
+All shared types live in `src/types.ts`. Add types there before writing implementations.
+
+### Adapter pattern
+
+Every primitive has an interface in `src/<module>/adapter.ts`. Filesystem implementations are the default. Future storage backends implement the same interface.
+
+| Module | Interface | Implementation |
+|--------|-----------|----------------|
+| memory | implicit in filesystem.ts | `src/memory/filesystem.ts` |
+| workflow | `WorkflowAdapter` | `src/workflow/filesystem.ts` |
+| tasks | `TaskAdapter` | `src/tasks/adapter.ts` (stub only) |
+| personas | `PersonaAdapter` | `src/personas/filesystem.ts` |
+| artifacts | `ArtifactAdapter` | `src/artifact/filesystem.ts` |
+
+### Source abstraction
+
+Config-style primitives (personas, skills, workflows, file-style dispatch configs) load through a pluggable `Source` interface in `src/sources/`:
+
+```typescript
+interface Source {
+  read(name: string): Promise<SourceRecord | undefined>  // text + locator
+  list(): Promise<string[]>
+}
+```
+
+Reference implementations:
+
+- `FlatFileSource(dir, ext)` — `<dir>/<name><ext>` layouts (personas, workflows)
+- `NestedFileSource(dir, filename)` — `<dir>/<name>/<filename>` layouts (skills)
+- `InMemorySource(records, tag)` — for tests and bake-in seed templates
+- `LayeredSource([liveSource, seedSource])` — first-wins read, union-dedupe list (the seed-then-emerge primitive)
+
+Per-primitive load functions (`loadPersonaFromSource`, `loadSkillFromSource`, `loadGraphFromSource`) accept any `Source` — Compass and other remote runtimes plug in their own (DB, HTTP) without touching spores.
+
+Data primitives (Memory, Artifacts, Tasks) have query semantics and live behind their own adapter shapes — `Source` is for config, not data.
+
+### CLI: two-word dispatch
+
+```
+agentic <noun> <verb> [args]    # preferred
+spores  <noun> <verb> [args]    # compatibility alias
+```
+
+Commands in `src/cli/commands/<noun>.ts`. Each command is a `Command` function exported as `<noun><Verb>Command`. The dispatch table is in `src/cli/main.ts`.
+
+Current command surface:
+- `agentic init` — scaffold `.agentic/` config
+- `agentic memory remember/recall/forget/dream/reinforce`
+- `agentic skill list/show/run`
+- `agentic workflow list/show/run/status`
+- `agentic persona list/view/activate`
+- `agentic artifact create/read/write/edit/inspect/list/finalize`
+
+### Skills on disk
+
+```
+~/.agentic/skills/<name>/skill.md    # global (user-level)
+.agentic/skills/<name>/skill.md      # project-level (wins on name conflict)
+# Legacy: ~/.spores/ and .spores/ are honoured when .agentic/ is absent
+```
+
+Frontmatter: `name`, `description`, `tags: [a, b, c]`
+Body: the skill content returned by `skill run` (pipe to an LLM).
+
+### Personas on disk
+
+```
+~/.agentic/personas/<name>.md        # global (user-level)
+.agentic/personas/<name>.md          # project-level (wins on name conflict)
+# Legacy: ~/.spores/ and .spores/ are honoured when .agentic/ is absent
+```
+
+Flat-file layout (unlike skills which use a directory per skill). Frontmatter: `name`, `description`, `memory_tags: [...]`, `skills: [...]`, optional `task_filter: { tags: [...], status: ready }` (nested, one level deep), optional `workflow: <graph-id>`. Body is markdown with `{{cwd}}`, `{{timestamp}}`, `{{hostname}}`, `{{git_branch}}` tokens that get substituted at `persona activate` time.
+
+**`view` vs `activate`** is load-bearing: `view` prints the raw file with literal tokens (for humans editing or reviewing); `activate` substitutes live situational facts (for piping into an LLM). Don't let them produce identical output.
+
+**One hat at a time.** Personas don't compose, stack, or inherit. To pivot, deactivate one and activate another. Runtime integration for applying persona bindings (using `memory_tags` as a recall filter, etc.) is **the caller's responsibility** — spores ships the metadata, the caller wires it. Descoped from v0.1 intentionally; expected to land after we have more signal from actual use.
+
+### Config resolution (four-tier)
+
+1. Hardcoded defaults in `config.ts`
+2. `~/.agentic/config.toml` — global user overrides (preferred); falls back to `~/.spores/config.toml`
+3. `.agentic/config.toml` — project overrides, wins over global (preferred); falls back to `.spores/config.toml`
+
+### Workflow runtime
+
+- `GraphDef` defines a digraph (nodes + edges with conditions)
+- `expandGraph` flattens nested subgraphs at register time — nesting is free
+- `Runtime` is a **state machine only** — it derives current state from `Run.history` (no `current_node` field). It does NOT schedule or evaluate `EvaluatorRef` conditions — that's the caller's job.
+- State is immutable: each transition appends to `history`
+
+### Dispatch
+
+Foundation only — the message shape and pure match logic. `Dispatch` carries `from`, `to`, `payload`, `timestamp`, plus optional `when` / `recurrence` delivery metadata. `DispatchFilter` is a declarative predicate over `from` / `to`; `matchDispatch(dispatch, filter)` returns boolean. `DispatchHandlerHooks` types `onRegister` / `onUnregister` for handler-level lifecycle setup (idempotency lives in the hook, not in spores).
+
+Runtimes own send/handle/cancel verbs, the actual transport, scheduling, and handler execution. The split — vocabulary in spores, execution in runtime — keeps spores callable from both long-running daemons and serverless deployments.
+
+See `PROJECTS/spores/DESIGN-runtime-description.md` for the full design conversation.
+
+### URI schemes
+
+`agentic://` is the preferred URI scheme for Agentic-owned compute. Branded type `AgenticUri = \`agentic://\${string}\`` in `types.ts`. Referenced from skill bodies, dispatched by the host runtime.
+
+`spores://` is the legacy scheme; `SporesUri` is preserved but deprecated. Use `AgenticUri` in new code.
+
+## Conventions
+
+- Test files colocated: `src/memory/filesystem.test.ts` next to `src/memory/filesystem.ts`
+- Test fixtures: inline (no separate fixtures dir)
+- IDs: ULIDs via monotonic factory (see tasks types)
+- Error handling: functions throw on unexpected errors; return `undefined` for "not found" cases (e.g. `loadSkill` returns `undefined` when skill doesn't exist)
+- No `console.log` in library code — CLI output goes through `output(ctx, data, formatter)` in `src/cli/main.ts`
+- **Descriptions are agent-facing activation triggers, not labels.** For both skills and personas, phrase `description` as "Activate when..." rather than "The X maintainer". `list` output is meant to function as a lookup table an agent scans to decide what to reach for — good triggers make the scan useful.
+
+### Artifact primitive
+
+Artifacts are versioned markdown blobs — the agent's durable output store for a project.
+
+**On-disk layout** (inside `.agentic/artifacts/`):
+
+```
+.agentic/artifacts/<ulid>/meta.json    — ArtifactRecord (type, title, version, finalized, tags, …)
+.agentic/artifacts/<ulid>/v1.md        — body at version 1
+.agentic/artifacts/<ulid>/v2.md        — body at version 2 (iterate write)
+```
+
+`body_ref` in `meta.json` is relative to `.agentic/artifacts/` — e.g. `"<id>/v2.md"`.
+Legacy: `.spores/artifacts/` is used when `.agentic/artifacts/` does not exist.
+
+**Write modes:**
+
+| Mode | Behavior |
+|------|----------|
+| `iterate` (default) | Bump version, write `v<n+1>.md`. Prior versions remain on disk. |
+| `replace` | Overwrite current `v<n>.md` in place. Version unchanged. |
+| `create` | Fail with "already exists". Used only for first-write semantics. |
+
+**Finalize semantics:** `artifact finalize <id>` sets `finalized=true` in `meta.json`. Finalized artifacts reject `write` and `edit`. `finalize` is idempotent — finalizing an already-finalized artifact is a no-op.
+
+**CLI worked example:**
+
+```bash
+# Create
+agentic artifact create brief "## Q2 Launch\n\nTBD." --title "Q2 Brief" --tags "q2,launch"
+
+# Read (pipe-friendly — raw body to stdout in human mode, JSON with --json)
+agentic artifact read 01JXYZ... | pbcopy
+
+# Iterate
+agentic artifact write 01JXYZ... "## Q2 Launch\n\nUpdated content." --mode iterate
+
+# Targeted edit
+agentic artifact edit 01JXYZ... --old "TBD." --new "Final copy."
+
+# Inspect metadata
+agentic artifact inspect 01JXYZ... --json
+
+# List with filter
+agentic artifact list --type brief --json
+
+# Finalize the artifact
+agentic artifact finalize 01JXYZ...
+```
+
+**Hook events:**
+
+| Event | Fired by |
+|-------|---------|
+| `artifact.created` | `artifact create` |
+| `artifact.written` | `artifact write`, `artifact edit` |
+| `artifact.finalized` | `artifact finalize` |
+
+Hook env vars: `AGENTIC_ARTIFACT_ID`, `AGENTIC_ARTIFACT_TYPE`, `AGENTIC_ARTIFACT_TITLE`, `AGENTIC_ARTIFACT_TAGS` (create); `AGENTIC_ARTIFACT_ID`, `AGENTIC_ARTIFACT_VERSION`, `AGENTIC_ARTIFACT_MODE` (written); `AGENTIC_ARTIFACT_ID`, `AGENTIC_ARTIFACT_FINAL_VERSION` (finalized). Legacy `SPORES_*` mirrors are set alongside for compatibility.
+
+**Dogfood hook:** `.agentic/hooks/artifact.written` — indexes the artifact reference into memory after every write so it's searchable via `agentic memory recall`.
+
+**Workflow → artifact worked example:**
+
+A workflow node declares its output type via `artifact.type` (from #50):
+
+```yaml
+nodes:
+  - id: write-brief
+    label: Write Q2 Brief
+    artifact:
+      type: brief
+      description: Q2 launch brief produced by the briefing workflow
+```
+
+After the workflow run completes and the node produces its output, the caller persists it:
+
+```bash
+# Node output lands in the transition; caller writes it as a named artifact
+BODY=$(agentic workflow status "$RUN_ID" --json | jq -r '.nodes[-1].artifact.content')
+ARTIFACT_ID=$(agentic artifact create brief "$BODY" --title "Q2 Brief" --tags "briefing,q2" --json | jq -r '.artifact.id')
+
+# artifact.written hook fires automatically → reference indexed into memory
+# Later retrieval
+agentic memory recall "Q2 Brief"
+```
+
+The dogfood hook wires the last step without any special-cased plumbing in spores itself.
+
+## What NOT to add
+
+- Sessions, webhooks, hosting — daemon-layer, not SPORES
+- Any adapter implementation for tasks until the interface is settled
+- Dependencies — keep `"dependencies": {}` clean
+- Per-module adapter interfaces for memory (memory follows the filesystem.ts shape directly, no separate adapter.ts)
diff --git a/CHANGELOG.md b/packages/agentic/CHANGELOG.md
similarity index 100%
rename from CHANGELOG.md
rename to packages/agentic/CHANGELOG.md
diff --git a/packages/agentic/README.md b/packages/agentic/README.md
new file mode 100644
index 0000000..46c3c33
--- /dev/null
+++ b/packages/agentic/README.md
@@ -0,0 +1,92 @@
+# @tnezdev/agentic
+
+> **Compatibility window.** This package was formerly `@tnezdev/spores`. New code should use `@tnezdev/agentic`, `agentic`, `.agentic/`, and `AGENTIC_*`. The `spores` CLI alias, `.spores/`, and `SPORES_*` remain supported during migration.
+
+**In-loop primitives for agents that run more than once.**
+
+Most agent tooling optimizes for the single prompt. Agentic optimizes for the agent that wakes up tomorrow — with memory, skills, workflows, tasks, and a persona that shape what it does next. It's the toolbelt inside the turn, not the runtime around it.
+
+Agentic is a primitive layer, not a host runtime. It gives your agent building blocks — store and recall memories, load skills, advance workflows, track tasks, activate personas — without owning scheduling, auth, transport, or execution. That's your daemon's job. Agentic just makes it easier to build.
+
+Zero production dependencies. Runs on Bun. Works from the CLI.
+
+## Why this exists
+
+Agents that run in production need more than a prompt and a model call. They need:
+
+- **Memory** that persists across sessions — not just a growing context window.
+- **Skills** that are loaded on demand, not baked into a system prompt.
+- **Workflows** that define multi-step processes with real state machines, not ad-hoc prompt chains.
+- **Tasks** that track commitments inside the loop, not just on an external board.
+- **Personas** that activate a coherent hat — memory tags, skill set, task filter, and rendered instructions — in one declaration.
+
+Agentic ships all five as filesystem-backed primitives your agent can reach for inside any turn. No hosting, no webhooks, no session layer required.
+
+## Install
+
+```bash
+npm install @tnezdev/agentic
+# or: bun add @tnezdev/agentic
+```
+
+## Quick start
+
+```bash
+# Scaffold .agentic/ directory in your project
+agentic init
+
+# Store a memory
+agentic memory remember "always emit types from the public API" --tags style,api
+
+# Load a skill
+agentic skill list
+agentic skill run release-check | llm
+
+# Track a task
+agentic task add "update CHANGELOG before tagging"
+agentic task next
+
+# Activate a persona (pipe into your LLM as system prompt)
+agentic persona list
+agentic persona activate spores-maintainer | llm --system -
+```
+
+## Primitives
+
+| Primitive | What it does |
+|-----------|-------------|
+| **Memory** | Tiered store (L1/L2/L3) with recall, reinforce, and dream consolidation |
+| **Skills** | Load `.md` skill files and pipe their content into an LLM |
+| **Workflow** | Directed-graph runtime — register a graph, create runs, advance node state |
+| **Tasks** | ULID-keyed task queue with status transitions and annotations |
+| **Persona** | Activate a "hat" at the start of a turn: memory tags, skills, task filter, workflow, and a rendered body with live situational facts |
+
+All five primitives read from `.agentic/` in your project root, with optional global overrides from `~/.agentic/`. Legacy `.spores/` and `~/.spores/` paths are honoured during the compatibility window when `.agentic/` is absent.
+
+## What Agentic is not
+
+Agentic is intentionally scoped. It does not ship:
+
+- Hosted sessions or transcript storage
+- Webhook receivers or background schedulers
+- Provider clients or credential management
+- Approval inboxes or auth flows
+- Platform-specific task adapters in core
+
+Those are host-runtime responsibilities. Agentic provides the vocabulary and primitives that make hosts easier to build, test, and reason about. See [`docs/framework-boundaries.md`](https://github.com/tnezdev/agentic/blob/main/docs/framework-boundaries.md) for the full boundary design.
+
+## Flags
+
+```
+--json          Output as JSON (machine-readable, no truncation)
+--wide          Disable column truncation in list output
+--base-dir      Override working directory
+```
+
+## Working example
+
+The `.agentic/` directory in this repo is the dogfood — a persona, a skill, a workflow, tasks, and memories that are used to build this package itself. Read [`.agentic/ONRAMP.md`](https://github.com/tnezdev/agentic/blob/main/.agentic/ONRAMP.md) for a tour.
+
+## Architecture
+
+See [AGENTS.md](AGENTS.md) for the full architecture, on-disk layout, and conventions for agent sessions.
diff --git a/packages/agentic/package.json b/packages/agentic/package.json
new file mode 100644
index 0000000..62888ab
--- /dev/null
+++ b/packages/agentic/package.json
@@ -0,0 +1,46 @@
+{
+  "name": "@tnezdev/agentic",
+  "version": "0.5.0",
+  "description": "Executable toolbelt for agent self-improvement",
+  "license": "MIT",
+  "type": "module",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/tnezdev/agentic.git",
+    "directory": "packages/agentic"
+  },
+  "author": "Travis Nesland",
+  "publishConfig": {
+    "access": "public"
+  },
+  "main": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "default": "./dist/index.js"
+    }
+  },
+  "bin": {
+    "agentic": "dist/cli/main.js",
+    "spores": "dist/cli/main.js"
+  },
+  "files": [
+    "dist/**/*.js",
+    "dist/**/*.d.ts",
+    "dist/**/*.map",
+    "AGENTS.md",
+    "CHANGELOG.md"
+  ],
+  "scripts": {
+    "build": "tsc -p tsconfig.build.json && chmod 755 dist/cli/main.js",
+    "typecheck": "tsc --noEmit -p tsconfig.json",
+    "test": "bun test",
+    "prepack": "bun run build"
+  },
+  "devDependencies": {
+    "@cloudflare/workers-types": "^4.20260426.1",
+    "@types/bun": "^1.3.11",
+    "typescript": "^5.7.0"
+  }
+}
diff --git a/src/artifact/adapter.ts b/packages/agentic/src/artifact/adapter.ts
similarity index 100%
rename from src/artifact/adapter.ts
rename to packages/agentic/src/artifact/adapter.ts
diff --git a/src/artifact/filesystem.test.ts b/packages/agentic/src/artifact/filesystem.test.ts
similarity index 100%
rename from src/artifact/filesystem.test.ts
rename to packages/agentic/src/artifact/filesystem.test.ts
diff --git a/src/artifact/filesystem.ts b/packages/agentic/src/artifact/filesystem.ts
similarity index 100%
rename from src/artifact/filesystem.ts
rename to packages/agentic/src/artifact/filesystem.ts
diff --git a/src/capability/catalog.test.ts b/packages/agentic/src/capability/catalog.test.ts
similarity index 100%
rename from src/capability/catalog.test.ts
rename to packages/agentic/src/capability/catalog.test.ts
diff --git a/src/capability/filesystem.test.ts b/packages/agentic/src/capability/filesystem.test.ts
similarity index 100%
rename from src/capability/filesystem.test.ts
rename to packages/agentic/src/capability/filesystem.test.ts
diff --git a/src/capability/filesystem.ts b/packages/agentic/src/capability/filesystem.ts
similarity index 100%
rename from src/capability/filesystem.ts
rename to packages/agentic/src/capability/filesystem.ts
diff --git a/src/capability/fixtures.test.ts b/packages/agentic/src/capability/fixtures.test.ts
similarity index 100%
rename from src/capability/fixtures.test.ts
rename to packages/agentic/src/capability/fixtures.test.ts
diff --git a/src/capability/fixtures/communication.place_call.json b/packages/agentic/src/capability/fixtures/communication.place_call.json
similarity index 100%
rename from src/capability/fixtures/communication.place_call.json
rename to packages/agentic/src/capability/fixtures/communication.place_call.json
diff --git a/src/capability/fixtures/document.create_slide_deck.json b/packages/agentic/src/capability/fixtures/document.create_slide_deck.json
similarity index 100%
rename from src/capability/fixtures/document.create_slide_deck.json
rename to packages/agentic/src/capability/fixtures/document.create_slide_deck.json
diff --git a/src/capability/fixtures/issue_tracker.create_issue.json b/packages/agentic/src/capability/fixtures/issue_tracker.create_issue.json
similarity index 100%
rename from src/capability/fixtures/issue_tracker.create_issue.json
rename to packages/agentic/src/capability/fixtures/issue_tracker.create_issue.json
diff --git a/src/capability/fixtures/issue_tracker.list_issues.json b/packages/agentic/src/capability/fixtures/issue_tracker.list_issues.json
similarity index 100%
rename from src/capability/fixtures/issue_tracker.list_issues.json
rename to packages/agentic/src/capability/fixtures/issue_tracker.list_issues.json
diff --git a/src/capability/fixtures/web.search.json b/packages/agentic/src/capability/fixtures/web.search.json
similarity index 100%
rename from src/capability/fixtures/web.search.json
rename to packages/agentic/src/capability/fixtures/web.search.json
diff --git a/src/capability/helpers.test.ts b/packages/agentic/src/capability/helpers.test.ts
similarity index 100%
rename from src/capability/helpers.test.ts
rename to packages/agentic/src/capability/helpers.test.ts
diff --git a/src/capability/helpers.ts b/packages/agentic/src/capability/helpers.ts
similarity index 100%
rename from src/capability/helpers.ts
rename to packages/agentic/src/capability/helpers.ts
diff --git a/src/cli/commands/artifact.test.ts b/packages/agentic/src/cli/commands/artifact.test.ts
similarity index 100%
rename from src/cli/commands/artifact.test.ts
rename to packages/agentic/src/cli/commands/artifact.test.ts
diff --git a/src/cli/commands/artifact.ts b/packages/agentic/src/cli/commands/artifact.ts
similarity index 100%
rename from src/cli/commands/artifact.ts
rename to packages/agentic/src/cli/commands/artifact.ts
diff --git a/src/cli/commands/capability.test.ts b/packages/agentic/src/cli/commands/capability.test.ts
similarity index 100%
rename from src/cli/commands/capability.test.ts
rename to packages/agentic/src/cli/commands/capability.test.ts
diff --git a/src/cli/commands/capability.ts b/packages/agentic/src/cli/commands/capability.ts
similarity index 100%
rename from src/cli/commands/capability.ts
rename to packages/agentic/src/cli/commands/capability.ts
diff --git a/src/cli/commands/init.ts b/packages/agentic/src/cli/commands/init.ts
similarity index 100%
rename from src/cli/commands/init.ts
rename to packages/agentic/src/cli/commands/init.ts
diff --git a/src/cli/commands/memory.test.ts b/packages/agentic/src/cli/commands/memory.test.ts
similarity index 100%
rename from src/cli/commands/memory.test.ts
rename to packages/agentic/src/cli/commands/memory.test.ts
diff --git a/src/cli/commands/memory.ts b/packages/agentic/src/cli/commands/memory.ts
similarity index 100%
rename from src/cli/commands/memory.ts
rename to packages/agentic/src/cli/commands/memory.ts
diff --git a/src/cli/commands/persona.ts b/packages/agentic/src/cli/commands/persona.ts
similarity index 100%
rename from src/cli/commands/persona.ts
rename to packages/agentic/src/cli/commands/persona.ts
diff --git a/src/cli/commands/skill.test.ts b/packages/agentic/src/cli/commands/skill.test.ts
similarity index 100%
rename from src/cli/commands/skill.test.ts
rename to packages/agentic/src/cli/commands/skill.test.ts
diff --git a/src/cli/commands/skill.ts b/packages/agentic/src/cli/commands/skill.ts
similarity index 100%
rename from src/cli/commands/skill.ts
rename to packages/agentic/src/cli/commands/skill.ts
diff --git a/src/cli/commands/task.test.ts b/packages/agentic/src/cli/commands/task.test.ts
similarity index 100%
rename from src/cli/commands/task.test.ts
rename to packages/agentic/src/cli/commands/task.test.ts
diff --git a/src/cli/commands/task.ts b/packages/agentic/src/cli/commands/task.ts
similarity index 100%
rename from src/cli/commands/task.ts
rename to packages/agentic/src/cli/commands/task.ts
diff --git a/src/cli/commands/wake.test.ts b/packages/agentic/src/cli/commands/wake.test.ts
similarity index 100%
rename from src/cli/commands/wake.test.ts
rename to packages/agentic/src/cli/commands/wake.test.ts
diff --git a/src/cli/commands/wake.ts b/packages/agentic/src/cli/commands/wake.ts
similarity index 100%
rename from src/cli/commands/wake.ts
rename to packages/agentic/src/cli/commands/wake.ts
diff --git a/src/cli/commands/workflow.test.ts b/packages/agentic/src/cli/commands/workflow.test.ts
similarity index 100%
rename from src/cli/commands/workflow.test.ts
rename to packages/agentic/src/cli/commands/workflow.test.ts
diff --git a/src/cli/commands/workflow.ts b/packages/agentic/src/cli/commands/workflow.ts
similarity index 100%
rename from src/cli/commands/workflow.ts
rename to packages/agentic/src/cli/commands/workflow.ts
diff --git a/src/cli/context.ts b/packages/agentic/src/cli/context.ts
similarity index 100%
rename from src/cli/context.ts
rename to packages/agentic/src/cli/context.ts
diff --git a/src/cli/format.ts b/packages/agentic/src/cli/format.ts
similarity index 100%
rename from src/cli/format.ts
rename to packages/agentic/src/cli/format.ts
diff --git a/src/cli/main.test.ts b/packages/agentic/src/cli/main.test.ts
similarity index 100%
rename from src/cli/main.test.ts
rename to packages/agentic/src/cli/main.test.ts
diff --git a/src/cli/main.ts b/packages/agentic/src/cli/main.ts
similarity index 100%
rename from src/cli/main.ts
rename to packages/agentic/src/cli/main.ts
diff --git a/src/cli/output.ts b/packages/agentic/src/cli/output.ts
similarity index 100%
rename from src/cli/output.ts
rename to packages/agentic/src/cli/output.ts
diff --git a/src/config.test.ts b/packages/agentic/src/config.test.ts
similarity index 100%
rename from src/config.test.ts
rename to packages/agentic/src/config.test.ts
diff --git a/src/config.ts b/packages/agentic/src/config.ts
similarity index 100%
rename from src/config.ts
rename to packages/agentic/src/config.ts
diff --git a/src/dispatch/match.test.ts b/packages/agentic/src/dispatch/match.test.ts
similarity index 100%
rename from src/dispatch/match.test.ts
rename to packages/agentic/src/dispatch/match.test.ts
diff --git a/src/dispatch/match.ts b/packages/agentic/src/dispatch/match.ts
similarity index 100%
rename from src/dispatch/match.ts
rename to packages/agentic/src/dispatch/match.ts
diff --git a/src/hooks/fire.test.ts b/packages/agentic/src/hooks/fire.test.ts
similarity index 100%
rename from src/hooks/fire.test.ts
rename to packages/agentic/src/hooks/fire.test.ts
diff --git a/src/hooks/fire.ts b/packages/agentic/src/hooks/fire.ts
similarity index 100%
rename from src/hooks/fire.ts
rename to packages/agentic/src/hooks/fire.ts
diff --git a/src/index.ts b/packages/agentic/src/index.ts
similarity index 100%
rename from src/index.ts
rename to packages/agentic/src/index.ts
diff --git a/src/lifecycle-events.test.ts b/packages/agentic/src/lifecycle-events.test.ts
similarity index 100%
rename from src/lifecycle-events.test.ts
rename to packages/agentic/src/lifecycle-events.test.ts
diff --git a/src/memory/adapter.ts b/packages/agentic/src/memory/adapter.ts
similarity index 100%
rename from src/memory/adapter.ts
rename to packages/agentic/src/memory/adapter.ts
diff --git a/src/memory/filesystem.test.ts b/packages/agentic/src/memory/filesystem.test.ts
similarity index 100%
rename from src/memory/filesystem.test.ts
rename to packages/agentic/src/memory/filesystem.test.ts
diff --git a/src/memory/filesystem.ts b/packages/agentic/src/memory/filesystem.ts
similarity index 100%
rename from src/memory/filesystem.ts
rename to packages/agentic/src/memory/filesystem.ts
diff --git a/src/personas/activate.test.ts b/packages/agentic/src/personas/activate.test.ts
similarity index 100%
rename from src/personas/activate.test.ts
rename to packages/agentic/src/personas/activate.test.ts
diff --git a/src/personas/activate.ts b/packages/agentic/src/personas/activate.ts
similarity index 100%
rename from src/personas/activate.ts
rename to packages/agentic/src/personas/activate.ts
diff --git a/src/personas/adapter.ts b/packages/agentic/src/personas/adapter.ts
similarity index 100%
rename from src/personas/adapter.ts
rename to packages/agentic/src/personas/adapter.ts
diff --git a/src/personas/filesystem.test.ts b/packages/agentic/src/personas/filesystem.test.ts
similarity index 100%
rename from src/personas/filesystem.test.ts
rename to packages/agentic/src/personas/filesystem.test.ts
diff --git a/src/personas/filesystem.ts b/packages/agentic/src/personas/filesystem.ts
similarity index 100%
rename from src/personas/filesystem.ts
rename to packages/agentic/src/personas/filesystem.ts
diff --git a/src/personas/situational.test.ts b/packages/agentic/src/personas/situational.test.ts
similarity index 100%
rename from src/personas/situational.test.ts
rename to packages/agentic/src/personas/situational.test.ts
diff --git a/src/personas/situational.ts b/packages/agentic/src/personas/situational.ts
similarity index 100%
rename from src/personas/situational.ts
rename to packages/agentic/src/personas/situational.ts
diff --git a/src/resolve-dir.ts b/packages/agentic/src/resolve-dir.ts
similarity index 100%
rename from src/resolve-dir.ts
rename to packages/agentic/src/resolve-dir.ts
diff --git a/src/skills/filesystem.test.ts b/packages/agentic/src/skills/filesystem.test.ts
similarity index 100%
rename from src/skills/filesystem.test.ts
rename to packages/agentic/src/skills/filesystem.test.ts
diff --git a/src/skills/filesystem.ts b/packages/agentic/src/skills/filesystem.ts
similarity index 100%
rename from src/skills/filesystem.ts
rename to packages/agentic/src/skills/filesystem.ts
diff --git a/src/sources/flat-file.test.ts b/packages/agentic/src/sources/flat-file.test.ts
similarity index 100%
rename from src/sources/flat-file.test.ts
rename to packages/agentic/src/sources/flat-file.test.ts
diff --git a/src/sources/flat-file.ts b/packages/agentic/src/sources/flat-file.ts
similarity index 100%
rename from src/sources/flat-file.ts
rename to packages/agentic/src/sources/flat-file.ts
diff --git a/src/sources/http.test.ts b/packages/agentic/src/sources/http.test.ts
similarity index 100%
rename from src/sources/http.test.ts
rename to packages/agentic/src/sources/http.test.ts
diff --git a/src/sources/http.ts b/packages/agentic/src/sources/http.ts
similarity index 100%
rename from src/sources/http.ts
rename to packages/agentic/src/sources/http.ts
diff --git a/src/sources/in-memory.test.ts b/packages/agentic/src/sources/in-memory.test.ts
similarity index 100%
rename from src/sources/in-memory.test.ts
rename to packages/agentic/src/sources/in-memory.test.ts
diff --git a/src/sources/in-memory.ts b/packages/agentic/src/sources/in-memory.ts
similarity index 100%
rename from src/sources/in-memory.ts
rename to packages/agentic/src/sources/in-memory.ts
diff --git a/src/sources/kv.test.ts b/packages/agentic/src/sources/kv.test.ts
similarity index 100%
rename from src/sources/kv.test.ts
rename to packages/agentic/src/sources/kv.test.ts
diff --git a/src/sources/kv.ts b/packages/agentic/src/sources/kv.ts
similarity index 100%
rename from src/sources/kv.ts
rename to packages/agentic/src/sources/kv.ts
diff --git a/src/sources/layered.test.ts b/packages/agentic/src/sources/layered.test.ts
similarity index 100%
rename from src/sources/layered.test.ts
rename to packages/agentic/src/sources/layered.test.ts
diff --git a/src/sources/layered.ts b/packages/agentic/src/sources/layered.ts
similarity index 100%
rename from src/sources/layered.ts
rename to packages/agentic/src/sources/layered.ts
diff --git a/src/sources/nested-file.test.ts b/packages/agentic/src/sources/nested-file.test.ts
similarity index 100%
rename from src/sources/nested-file.test.ts
rename to packages/agentic/src/sources/nested-file.test.ts
diff --git a/src/sources/nested-file.ts b/packages/agentic/src/sources/nested-file.ts
similarity index 100%
rename from src/sources/nested-file.ts
rename to packages/agentic/src/sources/nested-file.ts
diff --git a/src/sources/r2.test.ts b/packages/agentic/src/sources/r2.test.ts
similarity index 100%
rename from src/sources/r2.test.ts
rename to packages/agentic/src/sources/r2.test.ts
diff --git a/src/sources/r2.ts b/packages/agentic/src/sources/r2.ts
similarity index 100%
rename from src/sources/r2.ts
rename to packages/agentic/src/sources/r2.ts
diff --git a/src/sources/source.ts b/packages/agentic/src/sources/source.ts
similarity index 100%
rename from src/sources/source.ts
rename to packages/agentic/src/sources/source.ts
diff --git a/src/tasks/adapter.ts b/packages/agentic/src/tasks/adapter.ts
similarity index 100%
rename from src/tasks/adapter.ts
rename to packages/agentic/src/tasks/adapter.ts
diff --git a/src/tasks/filesystem.test.ts b/packages/agentic/src/tasks/filesystem.test.ts
similarity index 100%
rename from src/tasks/filesystem.test.ts
rename to packages/agentic/src/tasks/filesystem.test.ts
diff --git a/src/tasks/filesystem.ts b/packages/agentic/src/tasks/filesystem.ts
similarity index 100%
rename from src/tasks/filesystem.ts
rename to packages/agentic/src/tasks/filesystem.ts
diff --git a/src/types.ts b/packages/agentic/src/types.ts
similarity index 100%
rename from src/types.ts
rename to packages/agentic/src/types.ts
diff --git a/src/wake/resolve.test.ts b/packages/agentic/src/wake/resolve.test.ts
similarity index 100%
rename from src/wake/resolve.test.ts
rename to packages/agentic/src/wake/resolve.test.ts
diff --git a/src/wake/resolve.ts b/packages/agentic/src/wake/resolve.ts
similarity index 100%
rename from src/wake/resolve.ts
rename to packages/agentic/src/wake/resolve.ts
diff --git a/src/workflow/adapter.ts b/packages/agentic/src/workflow/adapter.ts
similarity index 100%
rename from src/workflow/adapter.ts
rename to packages/agentic/src/workflow/adapter.ts
diff --git a/src/workflow/expand.test.ts b/packages/agentic/src/workflow/expand.test.ts
similarity index 100%
rename from src/workflow/expand.test.ts
rename to packages/agentic/src/workflow/expand.test.ts
diff --git a/src/workflow/expand.ts b/packages/agentic/src/workflow/expand.ts
similarity index 100%
rename from src/workflow/expand.ts
rename to packages/agentic/src/workflow/expand.ts
diff --git a/src/workflow/filesystem.test.ts b/packages/agentic/src/workflow/filesystem.test.ts
similarity index 100%
rename from src/workflow/filesystem.test.ts
rename to packages/agentic/src/workflow/filesystem.test.ts
diff --git a/src/workflow/filesystem.ts b/packages/agentic/src/workflow/filesystem.ts
similarity index 100%
rename from src/workflow/filesystem.ts
rename to packages/agentic/src/workflow/filesystem.ts
diff --git a/src/workflow/parse.test.ts b/packages/agentic/src/workflow/parse.test.ts
similarity index 100%
rename from src/workflow/parse.test.ts
rename to packages/agentic/src/workflow/parse.test.ts
diff --git a/src/workflow/parse.ts b/packages/agentic/src/workflow/parse.ts
similarity index 100%
rename from src/workflow/parse.ts
rename to packages/agentic/src/workflow/parse.ts
diff --git a/src/workflow/runtime.test.ts b/packages/agentic/src/workflow/runtime.test.ts
similarity index 100%
rename from src/workflow/runtime.test.ts
rename to packages/agentic/src/workflow/runtime.test.ts
diff --git a/src/workflow/runtime.ts b/packages/agentic/src/workflow/runtime.ts
similarity index 100%
rename from src/workflow/runtime.ts
rename to packages/agentic/src/workflow/runtime.ts
diff --git a/src/workflow/yaml.test.ts b/packages/agentic/src/workflow/yaml.test.ts
similarity index 100%
rename from src/workflow/yaml.test.ts
rename to packages/agentic/src/workflow/yaml.test.ts
diff --git a/src/workflow/yaml.ts b/packages/agentic/src/workflow/yaml.ts
similarity index 100%
rename from src/workflow/yaml.ts
rename to packages/agentic/src/workflow/yaml.ts
diff --git a/tsconfig.build.json b/packages/agentic/tsconfig.build.json
similarity index 100%
rename from tsconfig.build.json
rename to packages/agentic/tsconfig.build.json
diff --git a/packages/agentic/tsconfig.json b/packages/agentic/tsconfig.json
new file mode 100644
index 0000000..52d43ea
--- /dev/null
+++ b/packages/agentic/tsconfig.json
@@ -0,0 +1,4 @@
+{
+  "extends": "../../tsconfig.json",
+  "include": ["src/**/*"]
+}
diff --git a/scripts/smoke-test-node.sh b/scripts/smoke-test-node.sh
index 1b49427..3d18e2e 100755
--- a/scripts/smoke-test-node.sh
+++ b/scripts/smoke-test-node.sh
@@ -17,11 +17,12 @@
 set -euo pipefail
 
 REPO_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
+PACKAGE_DIR="$REPO_ROOT/packages/agentic"
 TMPDIR_BASE="$(mktemp -d)"
 trap 'rm -rf "$TMPDIR_BASE"' EXIT
 
 echo "==> Packing tarball (runs prepack → bun run build)..."
-TARBALL=$(cd "$REPO_ROOT" && npm pack --silent --pack-destination "$TMPDIR_BASE" 2>/dev/null)
+TARBALL=$(cd "$PACKAGE_DIR" && npm pack --silent --pack-destination "$TMPDIR_BASE" 2>/dev/null)
 TARBALL_PATH="$TMPDIR_BASE/$TARBALL"
 
 if [ ! -f "$TARBALL_PATH" ]; then
diff --git a/scripts/smoke-test.sh b/scripts/smoke-test.sh
index 2bcef06..842202c 100755
--- a/scripts/smoke-test.sh
+++ b/scripts/smoke-test.sh
@@ -12,11 +12,12 @@
 set -euo pipefail
 
 REPO_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
+PACKAGE_DIR="$REPO_ROOT/packages/agentic"
 TMPDIR_BASE="$(mktemp -d)"
 trap 'rm -rf "$TMPDIR_BASE"' EXIT
 
 echo "==> Packing tarball (runs prepack → bun run build)..."
-TARBALL=$(cd "$REPO_ROOT" && npm pack --silent --pack-destination "$TMPDIR_BASE" 2>/dev/null)
+TARBALL=$(cd "$PACKAGE_DIR" && npm pack --silent --pack-destination "$TMPDIR_BASE" 2>/dev/null)
 TARBALL_PATH="$TMPDIR_BASE/$TARBALL"
 
 if [ ! -f "$TARBALL_PATH" ]; then
diff --git a/tsconfig.json b/tsconfig.json
index 4d6e073..59dff82 100644
--- a/tsconfig.json
+++ b/tsconfig.json
@@ -15,6 +15,5 @@
     "lib": ["ESNext"],
     "noEmit": true,
     "types": ["bun-types"]
-  },
-  "include": ["src/**/*"]
+  }
 }