@manifesto-ai/skills 0.2.2 → 0.3.0

package/README.md

@@ -1,7 +1,20 @@
 # @manifesto-ai/skills
 
+Standalone repository for the Manifesto integration skills pack.
+
 LLM knowledge pack for Manifesto.
 
+The packaged guidance is aligned to the current public seams:
+
+- base runtime via `@manifesto-ai/sdk`
+- continuity via `@manifesto-ai/lineage`
+- legitimacy via `@manifesto-ai/governance`
+- domain inspection and runtime debugging via `@manifesto-ai/studio-cli`
+- projection-first offline analysis via `@manifesto-ai/studio-core`
+- projected vs canonical snapshot boundaries
+
+This package is aimed at agents using Manifesto in apps, tools, and experiments, not at maintaining Manifesto core itself.
+
 ## Codex setup
 
 With npm:
@@ -20,6 +33,7 @@ pnpm exec manifesto-skills install-codex
 
 This installs the managed `manifesto` skill into `$CODEX_HOME/skills/manifesto` or `~/.codex/skills/manifesto`.
 The setup is explicit so package-manager `postinstall` approval policies do not block installation.
+Rerun the install command after upgrading `@manifesto-ai/skills` if you want the managed Codex skill refreshed to the latest seam knowledge.
 
 If package-manager exec is unavailable, run:
 
@@ -34,10 +48,15 @@ After installation, restart Codex.
 Add this to your `CLAUDE.md`:
 
 ```md
-See @node_modules/@manifesto-ai/skills/SKILL.md for Manifesto development rules.
+See @node_modules/@manifesto-ai/skills/SKILL.md for Manifesto integration guidance.
 ```
 
+Prefer wording like "integration guidance" when describing this package. It is meant for agents using Manifesto in an app, not for maintaining Manifesto core internals.
+
 ## Notes
 
 - This package does not auto-install Codex files from `postinstall`.
 - The installer refuses to overwrite an existing non-managed `manifesto` skill directory.
+- The managed install is safe to rerun when the package updates.
+- The managed install copies the skill and its knowledge files only. It does not vendor repo source docs or tracking metadata.
+- The installed knowledge is intended to be self-contained for normal integrations.

package/SKILL.md

@@ -1,19 +1,22 @@
 ---
 name: manifesto
-description: Use when working on Manifesto repositories, MEL flows, or Core/Host/SDK/World boundaries. Loads implementation-aligned architecture, patch, effect, and package guidance.
+description: Use when building with Manifesto runtimes, MEL flows, or SDK/Studio/Lineage/Governance seams. Focus on consumer-facing integration guidance, domain inspection, and runtime debugging.
 ---
 
-# Manifesto Skills v0.2.0
+# Manifesto Skills
 
-You are working on a Manifesto-based project. These rules are non-negotiable.
+You are integrating or extending a Manifesto-based system. Prefer the current public seams and keep your reasoning inside the installed skill knowledge by default.
 
 ## Scope Note
 
-This skills pack follows the current implementation in this repo.
+This skills pack is for agents using Manifesto, not for editing Manifesto internals by default.
 
-- `@manifesto-ai/world` is the canonical governed composition package and exact consumer-facing facade.
-- `@manifesto-ai/governance` and `@manifesto-ai/lineage` are implemented code packages and own their protocol-layer behavior.
-- For governed work, prefer top-level `@manifesto-ai/world` for composition and import split-native packages directly only when the task is intentionally limited to governance or lineage.
+- `@manifesto-ai/sdk` owns the base activation-first app path: `createManifesto(schema, effects) -> activate() -> dispatchAsync(intent)`.
+- `@manifesto-ai/lineage` and `@manifesto-ai/governance` are the active governed composition packages. The public governed direction is `createManifesto() -> withLineage() -> withGovernance() -> activate()`.
+- `getSnapshot()` is the normal app-facing read model. `getCanonicalSnapshot()` is the explicit substrate read for restore, seal-aware tooling, and deep debugging.
+- `@manifesto-ai/studio-cli` is the terminal inspection surface for findings, canonical snapshot debugging, trace replay, and transition graph projection.
+- `@manifesto-ai/studio-core` is the projection-first analysis layer for offline findings, graph inspection, and overlay-aware tooling.
+- The installed skill should be enough for normal app, tool, and experiment integrations. Do not assume repo-internal docs are available.
 
 ## Absolute Rules
 
@@ -25,9 +28,9 @@ This skills pack follows the current implementation in this repo.
 6. **Flows terminate.** No unbounded loops in Flow. Host controls iteration. All guards required for re-entry safety.
 7. **`$` is reserved.** `$host`, `$mel`, `$system` are platform namespaces. Never use `$` in domain identifiers.
 
-## Normative Hierarchy
+## Installed Knowledge Rule
 
-SPEC > FDR > ADR > Code > README. Never invent semantics not in SPEC.
+Prefer the installed knowledge files and the current public package seams. Do not assume repo-internal design documents are present.
 
 ## Task-Specific Knowledge
 
@@ -35,13 +38,15 @@ Load these before writing code in each area:
 
 | Task | Knowledge File |
 |------|---------------|
-| Understanding Core/Host/SDK/World boundaries | `@knowledge/architecture.md` |
+| Understanding base vs lineage vs governance runtime seams | `@knowledge/architecture.md` |
 | Writing MEL domain code | `@knowledge/mel-patterns.md` |
 | MEL complete function reference | `@knowledge/mel-reference.md` |
 | Implementing effect handlers | `@knowledge/effect-patterns.md` |
 | Working with state/patches | `@knowledge/patch-rules.md` |
 | Reviewing or debugging | `@knowledge/antipatterns.md` |
-| Looking up current specs and design docs | `@knowledge/spec-index.md` |
+| Choosing the right installed knowledge file | `@knowledge/spec-index.md` |
+| Inspecting or debugging a domain from the terminal | `@knowledge/packages/studio-cli.md` |
+| Understanding Studio projections and offline analysis | `@knowledge/packages/studio-core.md` |
 
 ## Package API Reference
 
@@ -50,11 +55,14 @@ Load when working with a specific package API:
 | Package | Knowledge File |
 |---------|---------------|
 | @manifesto-ai/sdk | `@knowledge/packages/sdk.md` |
+| @manifesto-ai/lineage | `@knowledge/packages/lineage.md` |
+| @manifesto-ai/governance | `@knowledge/packages/governance.md` |
 | @manifesto-ai/core | `@knowledge/packages/core.md` |
 | @manifesto-ai/host | `@knowledge/packages/host.md` |
-| @manifesto-ai/world | `@knowledge/packages/world.md` |
 | @manifesto-ai/compiler | `@knowledge/packages/compiler.md` |
 | @manifesto-ai/codegen | `@knowledge/packages/codegen.md` |
+| @manifesto-ai/studio-cli | `@knowledge/packages/studio-cli.md` |
+| @manifesto-ai/studio-core | `@knowledge/packages/studio-core.md` |
 
 ## Self-Check
 

package/claude-code/CLAUDE.md

@@ -1,6 +1,6 @@
-# Manifesto Development Rules
+# Manifesto Integration Rules
 
-This project uses the Manifesto framework. Read the rules below before writing any code.
+This project uses the Manifesto framework. Read the rules below before writing integration code.
 
 ## Entry Point
 
@@ -15,7 +15,7 @@ Load the relevant file BEFORE writing code:
 - Effect handler implementation: @../knowledge/effect-patterns.md
 - State mutations & patches: @../knowledge/patch-rules.md
 - Common mistakes to avoid: @../knowledge/antipatterns.md
-- SPEC document references: @../knowledge/spec-index.md
+- Knowledge index: @../knowledge/spec-index.md
 
 ## Quick Rules
 

package/knowledge/antipatterns.md

@@ -1,8 +1,5 @@
 # Antipatterns
 
-> Source: CLAUDE.md §10, Core FDR v2.0.0, Compiler SPEC v0.5.0 §8
-> Last synced: 2026-02-09
-
 If you see these patterns in code, REJECT the change.
 
 ## Constitutional Violations
@@ -22,7 +19,7 @@ async function executeEffect(req) {
 // Host MUST execute or report failure, never decide.
 ```
 
-Why: Host executes, Core computes. Decision-making is Core's job. [FDR-001]
+Why: Host executes, Core computes. Decision-making is Core's job.
 
 ### AP-002: Value Passing Outside Snapshot
 
@@ -37,7 +34,7 @@ core.compute(schema, snapshot, { ...intent, result });
 // Effect returns patches → Host applies → Next compute() reads from Snapshot
 ```
 
-Why: Snapshot is the sole communication medium. [FDR-002]
+Why: Snapshot is the sole communication medium.
 
 ### AP-003: Execution-Aware Core
 
@@ -51,7 +48,7 @@ if (effectExecutionSucceeded) { ... }
 if (snapshot.data.syncStatus === 'success') { ... }
 ```
 
-Why: Core is pure. It reads Snapshot, nothing else. [FDR-001]
+Why: Core is pure. It reads Snapshot, nothing else.
 
 ### AP-004: Hidden Continuation State
 
@@ -64,7 +61,7 @@ const pendingCallbacks = new Map();  // Hidden state!
 // CORRECT — All execution state in snapshot.system.pendingRequirements
 ```
 
-Why: There is no suspended execution context. [FDR-003]
+Why: There is no suspended execution context.
 
 ## Flow Violations
 
@@ -88,7 +85,7 @@ action increment() {
 }
 ```
 
-Why: Flows are re-evaluated on every `compute()` call. Guards prevent re-execution. [FDR-006]
+Why: Flows are re-evaluated on every `compute()` call. Guards prevent re-execution.
 
 ### AP-006: Unbounded Loop in Flow
 
@@ -99,7 +96,7 @@ Attempting Turing-complete loops in Flow.
 while eq(pending, true) { ... }
 ```
 
-Why: Flow must terminate in finite steps. Host controls iteration. [FDR-006]
+Why: Flow must terminate in finite steps. Host controls iteration.
 
 ### AP-007: Nested Effects
 
@@ -123,7 +120,7 @@ effect array.flatMap({ source: teams, select: $item.members, into: allMembers })
 effect array.filter({ source: allMembers, where: eq($item.active, true), into: activeMembers })
 ```
 
-Why: Effects are sequential declarations, not expressions. [Compiler SPEC §8.4]
+Why: Effects are sequential declarations, not expressions.
 
 ## State Violations
 
@@ -137,7 +134,7 @@ snapshot.data.count = 5;
 core.apply(schema, snapshot, [{ op: 'set', path: 'data.count', value: 5 }]);
 ```
 
-Why: Snapshots are immutable. All mutations via patches. [FDR-002]
+Why: Snapshots are immutable. All mutations via patches.
 
 ### AP-009: Throwing in Effect Handler
 
@@ -153,7 +150,7 @@ async function handler(params) {
 }
 ```
 
-Why: Errors are values in Snapshot, never thrown. [FDR-005]
+Why: Errors are values in Snapshot, never thrown.
 
 ## Type Violations
 
@@ -173,18 +170,18 @@ Why: User-facing APIs must be type-safe with zero string paths.
 
 ### AP-011: Authority Bypass
 
-Executing intents without World Protocol.
+Executing governed work without the current governance path.
 
 ```typescript
 // FORBIDDEN
 host.execute(snapshot, intent);  // Skips governance!
 
 // CORRECT
-world.submitProposal(actor, intentInstance);
-// Authority evaluates → approved intents go to Host
+const governed = withGovernance(withLineage(createManifesto(schema, effects), lineageConfig), governanceConfig).activate();
+await governed.proposeAsync(governed.createIntent(governed.MEL.actions.someAction));
 ```
 
-Why: All intents must pass through World Protocol for governance and auditability.
+Why: Governed work must pass through the current approval path to preserve legitimacy and auditability.
 
 ## MEL-Specific Mistakes
 
@@ -230,7 +227,7 @@ computed a = b
 computed b = a
 ```
 
-Computed dependencies MUST form a DAG. Cycles are rejected at validation time. [Core SPEC §14 V-002]
+Computed dependencies MUST form a DAG. Cycles are rejected at validation time.
 
 ## Quick Checklist
 

package/knowledge/architecture.md

@@ -1,122 +1,64 @@
-# Manifesto Architecture
+# Manifesto Runtime Map
 
-> Source: `packages/core/docs/core-SPEC.md`, `packages/host/docs/host-SPEC.md`, `packages/sdk/docs/sdk-SPEC-v2.0.0.md`, current `packages/world/src/*`
-> Last synced: 2026-03-31
+Use this file to choose the right public seam quickly.
 
-## Rules
+## Mental model
 
-> **R1**: Core computes, Host executes. These concerns never mix.
-> **R2**: Snapshot is the only medium of communication. If it's not in Snapshot, it doesn't exist.
-> **R3**: There is no suspended execution context. All continuity is expressed through Snapshot.
-> **R4**: Effects are declarations, not executions. Core declares; Host fulfills.
-> **R5**: If you need a value, read it from Snapshot. There is no other place.
+- Core computes meaning.
+- Host executes requirements and applies patches.
+- SDK exposes the base runtime for direct dispatch.
+- Lineage decorates the base runtime with continuity and sealing.
+- Governance decorates a lineage runtime with legitimacy and approval.
 
-## The Fundamental Equation
+## Runtime ladder
 
 ```typescript
-compute(schema, snapshot, intent, context) -> {
-  patches,
-  systemDelta,
-  trace,
-  status,
-}
+createManifesto(schema, effects).activate()
+  -> dispatchAsync(intent)
 ```
 
-- **Pure**: Same input must produce the same result.
-- **Total**: Business logic failures are reported as values, not thrown.
-- **Traceable**: Compute returns a trace graph for explainability.
-- **Resumable via Snapshot**: Host applies `patches` + `systemDelta`, then re-enters `compute()` with the new snapshot.
-
-## Runtime Paths
-
-Two practical paths exist in the current repo:
-
-```text
-SDK-style app
-  createManifesto()
-    -> Host
-    -> Core
-```
+Use this when you want the shortest path: compile/activate/dispatch/read.
 
 ```text
-Governed path
-  Actor submits IntentInstance
-    -> World
-    -> WorldExecutor
+Lineage runtime
+  createManifesto(schema, effects)
+    -> withLineage(...)
+    -> activate()
+    -> commitAsync(intent)
     -> Host
     -> Core
-    -> new World state + lineage records
-```
-
-## Package Sovereignty
-
-| Package | Responsibility | MUST NOT |
-|---------|---------------|----------|
-| **Core** | Pure computation, expression evaluation, flow interpretation, patch generation, validation, explanation | IO, wall-clock access, effect execution, Host/World policy |
-| **Host** | Effect execution, patch application, compute loop orchestration, requirement fulfillment | Compute semantic meaning, suppress declared effects, make governance decisions |
-| **World** | Proposal lifecycle, authority evaluation, lineage DAG, persistence, governance event emission | Import `@manifesto-ai/host` directly, compute semantic meaning, apply Core patches itself |
-| **SDK** | Public app entrypoint (`createManifesto`), typed effect registration, `dispatchAsync`, typed patch helpers, selected re-exports | Invent semantics outside Core/Host/World public contracts |
-
-## Current Governed Structure
-
-In this repo's implementation:
-
-- `@manifesto-ai/world` is the exact consumer-facing governed facade
-- `@manifesto-ai/governance` is the current governance protocol package
-- `@manifesto-ai/lineage` is the current continuity protocol package
-- top-level `@manifesto-ai/world` re-exports the split-native surfaces needed for explicit composition
-
-## Snapshot Structure
-
-```typescript
-type Snapshot = {
-  data: unknown;
-  computed: Record<string, unknown>;
-  system: {
-    status: "idle" | "computing" | "pending" | "error";
-    lastError: ErrorValue | null;
-    pendingRequirements: Requirement[];
-    currentAction: string | null;
-  };
-  input: unknown;
-  meta: {
-    version: number;
-    timestamp: number;
-    randomSeed: string;
-    schemaHash: string;
-  };
-};
+    -> seal into lineage
+    -> visible head advances
 ```
 
-## Compute / Apply Cycle
+Use this when execution must also become recorded continuity.
 
 ```text
-Host calls compute(schema, snapshot, intent, context)
-  -> Core returns patches + systemDelta + trace + status
-  -> Host applies patches
-  -> Host applies systemDelta
-  -> If status is pending, Host fulfills requirements and calls compute() again
+Governed runtime
+  createManifesto(schema, effects)
+    -> withLineage(...)
+    -> withGovernance(...)
+    -> activate()
+    -> proposeAsync(intent)
+    -> authority decision
+    -> approved commit
+    -> seal into lineage
+    -> visible governed history advances
 ```
 
-Each `compute()` call is complete and independent. Continuity lives in the snapshot, not in hidden runtime state.
+Use this when actions need legitimacy, approval, or proposal records.
 
-## Platform Namespaces
+## Snapshot boundary
 
-- `$host` is Host-owned internal state.
-- `$mel` is compiler-owned guard state.
-- `$system.*` values are surfaced in MEL and lowered through platform mechanics.
-- Domain schemas must not define `$`-prefixed fields.
+For application code, prefer the SDK projected snapshot from `getSnapshot()`. Escalate to `getCanonicalSnapshot()` only when you need substrate-level fields such as `pendingRequirements`, `currentAction`, or canonical metadata.
 
-## Why
+Practical rule:
 
-- **Determinism**: Core can be tested without mocks.
-- **Auditability**: World records proposal and decision lineage.
-- **Portability**: Host remains the execution seam.
-- **Clarity**: SDK is the public app-facing direct-dispatch layer, while World is the explicit governed composition layer around Host.
+- app/UI/agent reasoning: `getSnapshot()`
+- lineage restore / sealing / deep runtime tooling: `getCanonicalSnapshot()`
 
-## Cross-References
+## Quick reminders
 
-- MEL syntax: `@knowledge/mel-patterns.md`
-- Effect handlers: `@knowledge/effect-patterns.md`
-- Patch operations: `@knowledge/patch-rules.md`
-- World package API: `@knowledge/packages/world.md`
+- Effects are declarations from Core and IO adapters in Host handlers.
+- Snapshot is the only medium between turns and layers.
+- Only three patch ops exist: `set`, `unset`, `merge`.

package/knowledge/effect-patterns.md

@@ -1,8 +1,5 @@
 # Effect Patterns
 
-> Source: `packages/host/docs/host-SPEC.md`, `packages/sdk/src/types.ts`, `packages/sdk/src/create-manifesto.ts`
-> Last synced: 2026-03-28
-
 ## Rules
 
 > **R1**: Effect handlers return `Patch[]` and should not use exceptions as the business-logic result channel.
@@ -33,14 +30,11 @@ Host uses a different internal signature, but SDK adapts your handler automatica
 ```typescript
 import { createManifesto } from "@manifesto-ai/sdk";
 
-const app = createManifesto({
-  schema: domainSchema,
-  effects: {
-    "api.fetchUser": fetchUser,
-    "api.createTodo": createTodo,
-    "payment.process": processPayment,
-  },
-});
+const runtime = createManifesto(domainSchema, {
+  "api.fetchUser": fetchUser,
+  "api.createTodo": createTodo,
+  "payment.process": processPayment,
+}).activate();
 ```
 
 Effect names must match effect types declared in schema or MEL.

package/knowledge/llm/LLM-INDEX.md

@@ -1,82 +1,19 @@
 # Manifesto LLM Knowledge Pack Index
 
-Purpose: provide a stable map of sources that match the current repo layout.
+This installed pack is meant to be self-contained for normal integrations.
 
-## Normative hierarchy
-1) SPEC
-2) FDR
-3) ADR
-4) Guides / README
+## Default routing
 
-When implementation and future-design docs diverge, use the implemented package surface for coding tasks.
+- Runtime seam choice: `../architecture.md`
+- Base runtime API: `../packages/sdk.md`
+- Continuity and sealing: `../packages/lineage.md`
+- Legitimacy and approval: `../packages/governance.md`
+- Effect handlers: `../effect-patterns.md`
+- Patch and snapshot rules: `../patch-rules.md`
+- MEL generation patterns: `../mel-patterns.md`
+- MEL builtin summary: `../mel-reference.md`
+- Common mistakes: `../antipatterns.md`
 
-## Current code-writing sources
+## Working rule
 
-### Core
-- `packages/core/docs/core-SPEC.md`
-- `packages/core/docs/VERSION-INDEX.md`
-- `packages/core/src/index.ts`
-
-### Host
-- `packages/host/docs/host-SPEC.md`
-- `packages/host/docs/VERSION-INDEX.md`
-- `packages/host/src/index.ts`
-
-### World
-- `packages/world/docs/world-facade-spec-v1.0.0.md`
-- `packages/world/docs/VERSION-INDEX.md`
-- `packages/world/src/index.ts`
-- `packages/world/src/facade/index.ts`
-
-Note: top-level `@manifesto-ai/world` is the exact governed facade in the current repo.
-
-### Governance
-- `packages/governance/docs/governance-SPEC-1.0.0v.md`
-- `packages/governance/docs/VERSION-INDEX.md`
-- `packages/governance/src/index.ts`
-
-### Lineage
-- `packages/lineage/docs/lineage-SPEC-1.0.1v.md`
-- `packages/lineage/docs/VERSION-INDEX.md`
-- `packages/lineage/src/index.ts`
-
-### SDK
-- `packages/sdk/docs/sdk-SPEC-v2.0.0.md`
-- `packages/sdk/docs/VERSION-INDEX.md`
-- `packages/sdk/src/index.ts`
-- `packages/sdk/src/types.ts`
-- `packages/sdk/src/create-manifesto.ts`
-
-### Compiler
-- `packages/compiler/docs/SPEC-v0.7.0.md`
-- `packages/compiler/docs/VERSION-INDEX.md`
-- `packages/compiler/src/index.ts`
-
-### Codegen
-- `packages/codegen/docs/SPEC-v0.1.1.md`
-- `packages/codegen/docs/VERSION-INDEX.md`
-
-## Future-split / design-only references
-
-These are useful next-major drafts, but they are not the default coding target unless a task explicitly targets them:
-
-- `packages/sdk/docs/sdk-SPEC-v3.0.0-draft.md`
-- `packages/governance/docs/governance-SPEC-2.0.0v.md`
-- `packages/lineage/docs/lineage-SPEC-2.0.0v.md`
-- `packages/world/docs/world-facade-spec-v2.0.0.md`
-
-Use them for next-major direction, not as the primary basis for current code changes.
-
-## ADRs
-
-- ADR-002: `onceIntent` + `$mel` namespace
-- ADR-010: protocol-first SDK reconstruction
-- ADR-014: split world protocol
-
-## Legacy note
-
-Older application-layer docs and migration notes may still exist for historical context. They are not part of the current routing baseline for new code changes.
-
-## Archives
-
-Historical docs under `archive/` remain useful for archaeology, but they are not the primary source for new code unless a task explicitly targets historical behavior.
+Prefer the installed knowledge files first. Only reach outside the installed skill if the user explicitly asks for repo archaeology or maintainer-level implementation detail.

package/knowledge/llm/LLM-NORMS.md

@@ -1,17 +1,10 @@
 # Manifesto Normative Summary (LLM)
 
-This is a compact summary of high-salience rules. For exact wording, defer to the sources listed in `./LLM-INDEX.md`.
-
-## Document precedence
-
-- SPEC is authoritative.
-- FDR is rationale and must not contradict SPEC.
-- ADR is decision context and must not contradict SPEC or FDR.
-- Guides and READMEs are non-normative.
+This is a compact summary of high-salience rules for the installed skill.
 
 ## Core boundaries
 
-- Core computes meaning; Host executes effects; World governs proposals and lineage.
+- Core computes meaning; Host executes effects; Lineage preserves continuity; Governance governs legitimacy.
 - Core is pure and deterministic: same input, same output.
 - Snapshot is the sole communication medium between layers.
 - State changes are expressed as patches plus system transitions, not hidden mutable state.
@@ -42,12 +35,12 @@ This is a compact summary of high-salience rules. For exact wording, defer to th
 
 ## Current implementation note
 
-- In this repo, top-level `@manifesto-ai/world` is the exact consumer-facing governed facade.
-- `@manifesto-ai/governance` and `@manifesto-ai/lineage` are implemented code packages and own their protocol-layer behavior.
-- For consumer-facing governed work, prefer current `@manifesto-ai/world` exports and adapter subpaths. Import split-native packages directly only when the task is intentionally scoped to governance or lineage.
+- `@manifesto-ai/sdk` owns the activation-first base runtime.
+- `@manifesto-ai/lineage` and `@manifesto-ai/governance` are the active governed composition packages.
+- For governed work, prefer `createManifesto() -> withLineage() -> withGovernance() -> activate()`.
 
 ## Guidance for LLM use
 
-- Do not infer behavior not stated in SPEC, FDR, ADR, or current exported code.
-- When current and next-major docs differ, call that out explicitly and prefer the current exported surface unless the task targets a draft.
-- Prefer current package exports and source layout for implementation guidance.
+- Do not infer behavior not stated in the installed knowledge or current public package APIs.
+- Prefer the installed package notes over repo archaeology for normal integrations.
+- If something conflicts with the installed knowledge, prefer the current public seams: SDK, Lineage, Governance.

package/knowledge/mel-patterns.md

@@ -1,15 +1,12 @@
 # MEL Patterns
 
-> Source: Compiler SPEC v0.5.0, Compiler FDR v0.5.0
-> Last synced: 2026-02-09
-
 ## Rules
 
-> **R1**: MEL is pure and total. All expressions terminate, are deterministic, and never throw. [A1-A3]
-> **R2**: One pattern per concept. Function calls only, no method chaining. [A6]
-> **R3**: Every mutation (patch/effect) MUST be inside a guard (when/once/onceIntent). [A7]
-> **R4**: Effects are declarations, not executions. Effects are sequential, never nested. [A20]
-> **R5**: `$` is completely prohibited in user identifiers — not just at start, anywhere. [A28]
+> **R1**: MEL is pure and total. All expressions terminate, are deterministic, and never throw.
+> **R2**: One pattern per concept. Function calls only, no method chaining.
+> **R3**: Every mutation (patch/effect) MUST be inside a guard (when/once/onceIntent).
+> **R4**: Effects are declarations, not executions. Effects are sequential, never nested.
+> **R5**: `$` is completely prohibited in user identifiers — not just at start, anywhere.
 
 ## Domain Structure
 

package/knowledge/mel-reference.md

@@ -1,11 +1,6 @@
-# MEL Complete Reference
+# MEL Quick Reference
 
-> Source: docs/mel/REFERENCE.md
-> Last synced: 2026-03-24
-
-This is a pointer to the comprehensive MEL reference document. For the full content, see:
-
-**`docs/mel/REFERENCE.md`** (1700+ lines)
+This is the installed quick reference for MEL. Use it for normal generation work.
 
 ## Quick Summary
 

package/knowledge/packages/codegen.md

@@ -31,5 +31,4 @@ interface CodegenPlugin {
 
 ## Notes
 
-- Current docs live in `packages/codegen/docs/SPEC-v0.1.1.md`.
 - Keep output deterministic and plugin interactions side-effect free.

package/knowledge/packages/compiler.md

@@ -41,5 +41,5 @@ MEL text -> Lexer -> Parser -> Analyzer -> Generator -> Lowering
 
 ## Notes
 
-- Current spec baseline is `packages/compiler/docs/SPEC-v0.7.0.md`.
-- FDR baseline remains `packages/compiler/docs/FDR-v0.5.0.md`.
+- Prefer MEL source plus the compiler API surface shown here.
+- For normal integrations, you usually only need compile output or bundler wiring, not compiler internals.