5 changed files with 708 additions and 0 deletions
@@ -291,3 +291,101 @@ The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
  renders permission requests directly in the conversation stream for single-file
  operations. Users can allow/reject with single-key shortcuts (`a`/`A`/`r`/`R`),
  navigate with arrow keys, confirm with `Enter`, or press `v` to open the full
+
+## [3.1.0] — 2025-Q4
+
+### Added
+
+- **Actor Compiler** — GRAPH-type actor YAML files now compile into live
+  LangGraph `StateGraph` structures via `ActorCompiler`. Compilation validates
+  node references, detects intra-graph and cross-actor subgraph cycles, maps
+  node types to LangGraph `NodeType` values, and returns a `CompiledActor`
+  bundle with nodes, edges, entry point, and `CompilationMetadata`.
+
+- **`agents actor add --config actor.yaml`** — New CLI command loads and
+  registers custom actors from YAML files. The loader parses, validates via
+  Pydantic, enforces graph topology rules, and persists to the actor registry.
+  Supports `--update`, `--unsafe`, and `--set-default` flags.
+
+- **Actor YAML `version: "3"` schema** — Actor YAML files now declare
+  `version: "3"` and support `type: llm | tool | graph`. All fields are
+  validated via Pydantic with precise error messages including dotted field
+  paths.
+
+- **Tool Call Router** — `ToolCallRouter` normalizes OpenAI, Anthropic, and
+  LangChain tool call formats into `NormalizedToolCallResult` objects.
+  Supports single routing, batch routing, and streaming execution. Generates
+  stable deterministic tool call IDs from `plan_id + sequence`.
+
+- **MCP Adapter** — `MCPToolAdapter` connects to external MCP (Model Context
+  Protocol) tool servers, discovers tools, validates inputs against JSON
+  Schema, and bulk-registers tools into the `ToolRegistry`. Supports `stdio`,
+  `sse`, and `streamable-http` transports. Thread-safe via `threading.RLock`.
+
+- **Validation Runner** — Executes resource-attached validations before the
+  apply gate. Classifies validations as required (blocking) or informational
+  (non-blocking). `ApplyValidationSummary.all_required_passed` returns `False`
+  when zero validations were run (empty-run guard).
+
+- **Multi-file ChangeSet generation** — The execution actor now correctly
+  produces `ChangeSet` records with `added_files`, `modified_files`, and
+  `deleted_files` for multi-file generation scenarios.
+
+- **Skill registry and tool lifecycle CLI** — `agents skill add/list/show/remove`
+  commands manage named skill collections. Skills provide tool sets to actors
+  during plan execution.
+
+---
+
+## [3.0.0] — 2025-Q3
+
+### Added
+
+- **`agents action create --config action.yaml`** — Register reusable action
+  templates from YAML files. Actions are persisted to SQLite and include
+  `name`, `description`, `strategy_actor`, `execution_actor`,
+  `definition_of_done`, optional `arguments`, `invariants`, `inputs_schema`,
+  and `automation_profile`.
+
+- **`agents resource add git-checkout`** — Register git repositories as
+  resources in the resource registry. The `git-checkout` resource type is the
+  primary resource type for the git worktree sandbox.
+
+- **`agents project create` and `agents project link-resource`** — Create
+  projects and link resources to them. Projects group resources together and
+  provide context for plan execution.
+
+- **`agents plan use`** — Create plan records from action templates and
+  projects. Plans are persisted to SQLite with ULID identifiers and start in
+  the `strategize/queued` state.
+
+- **`agents plan execute <plan_id>`** — Run the Strategize and Execute phases
+  of the plan lifecycle. Invokes the actor-based LLM path and writes generated
+  file changes to the git worktree sandbox.
+
+- **`agents plan diff <plan_id>`** — Show pending sandbox changes as a unified
+  diff for review before applying.
+
+- **`agents plan apply <plan_id>`** — Merge sandbox changes into the real git
+  repository via `git merge` from the isolated worktree branch. Non-git
+  projects fall back to flat file copy.
+
+- **Git worktree sandbox** — Isolated working directories for plan execution.
+  Each plan gets its own branch (`cleveragents/plan-<plan_id>`) and a
+  temporary worktree. On apply, the sandbox branch is merged back into the
+  original branch and the worktree is cleaned up.
+
+- **Pydantic v2 domain models with `frozen=True`** — All domain models
+  (`ActionRecord`, `ResourceRecord`, `ProjectRecord`, `PlanRecord`,
+  `ChangeSet`, etc.) use Pydantic v2 with `frozen=True` for immutability
+  guarantees.
+
+- **SQLite persistence with Alembic migrations** — All records are persisted
+  to a local SQLite database. Schema migrations are managed by Alembic and
+  run automatically on first use.
+
+- **Plan state machine** — Plans follow a strict state machine:
+  `strategize/queued` → `strategize/in_progress` → `strategize/complete` →
+  `execute/queued` → `execute/in_progress` → `execute/complete` →
+  `apply/queued` → `apply/in_progress` → `applied`. Terminal states:
+  `applied`, `cancelled`, `errored`.
@@ -170,3 +170,41 @@ Set `CLEVERAGENTS_DEFAULT_PROVIDER` to pin the global provider (for example `exp
 - Built-in actors (`<provider>/<model>`) are immutable, custom actors must be named `local/<id>`, and the default actor cannot be removed. Use `--unsafe` when adding/updating configs marked unsafe; runtime only warns when invoking unsafe actors.
 - `CLEVERAGENTS_TESTING_USE_MOCK_AI=true` forces the in-repo mock provider so Behave/Robot suites never hit external APIs.
 - The full capability matrix (streaming, tool calls, JSON mode, etc.) is documented in `docs/reference/providers.md`.
+
+## Milestone History
+
+### v3.0.0 — M1: Minimal Local Source-Code Workflow
+
+The first minimally usable local-mode flow. Register an action from YAML,
+link a git repository resource to a project, and run a plan end-to-end.
+
+**Key capabilities:**
+- `agents action create --config action.yaml` — register action templates
+- `agents resource add git-checkout` — register git repository resources
+- `agents project create` / `agents project link-resource` — project management
+- `agents plan use` — create plan records with state machine transitions
+- `agents plan execute <plan_id>` — invoke actor-based LLM path
+- `agents plan diff <plan_id>` — review pending sandbox changes
+- `agents plan apply <plan_id>` — merge sandbox changes with git commit
+- Git worktree sandbox for isolated working directories
+- Pydantic v2 domain models with `frozen=True`
+- SQLite persistence with Alembic migrations
+
+See [`docs/modules/milestone-v3.0.0-local-workflow.md`](docs/modules/milestone-v3.0.0-local-workflow.md) for the full guide.
+
+### v3.1.0 — M2: Actor Compiler + Full LLM Integration
+
+Actor YAML files compile into live LangGraph graphs. Custom actors are fully
+operational with a normalized tool router, validation runner, and MCP adapter.
+
+**Key capabilities:**
+- Actor YAML files with `version: "3"`, `type: llm|tool|graph` — parse and validate via Pydantic
+- GRAPH-type actors compile into LangGraph `StateGraph` structures
+- `agents actor add --config actor.yaml` — load and register custom actors
+- Skill registry and tool lifecycle via CLI (`agents skill add/list/show/remove`)
+- MCP adapter discovers and connects to external tool servers
+- Tool call router normalizes OpenAI / Anthropic / LangChain formats
+- Validation runner executes required and informational validations
+- Multi-file generation produces correct `ChangeSet`
+
+See [`docs/modules/milestone-v3.1.0-actor-compiler.md`](docs/modules/milestone-v3.1.0-actor-compiler.md) for the full guide.
@@ -0,0 +1,265 @@
+# v3.0.0 — M1: Minimal Local Source-Code Workflow
+
+**Milestone:** v3.0.0 — M1 (CLOSED)
+**Released:** 2025-Q3
+
+This milestone delivered the first minimally usable local-mode flow for
+CleverAgents. A user can register an action from YAML, link a git repository
+resource to a project, and run a plan end-to-end — from creation through
+execution, diff review, and final apply — using a sandboxed workspace with
+tool-based change capture.
+
+---
+
+## Overview
+
+The M1 milestone established the foundational plan lifecycle:
+
+```
+agents action create  →  agents resource add  →  agents project create
+        │                                                │
+        └──────────────────────────────────────────────►│
+                                                         │
+                                               agents plan use
+                                                         │
+                                               agents plan execute
+                                                         │
+                                               agents plan diff
+                                                         │
+                                               agents plan apply
+```
+
+All state is persisted to SQLite via Alembic-managed migrations. Domain
+models use Pydantic v2 with `frozen=True` for immutability guarantees.
+
+---
+
+## CLI Commands Introduced
+
+### `agents action create`
+
+Register a reusable action template from a YAML configuration file.
+
+```bash
+agents action create --config action.yaml
+```
+
+**Minimal `action.yaml`:**
+
+```yaml
+name: local/my-action
+description: Describe what this action does
+strategy_actor: openai/gpt-4
+execution_actor: openai/gpt-4
+definition_of_done: |
+  The task is complete when all acceptance criteria are met.
+```
+
+Actions are persisted to SQLite and can be listed with `agents action list`.
+See [Action CLI Reference](../reference/action_cli.md) for the full field
+listing and options.
+
+---
+
+### `agents resource add git-checkout`
+
+Register a git repository as a resource in the resource registry.
+
+```bash
+agents resource add git-checkout local/my-repo --path /home/user/my-repo
+```
+
+The `git-checkout` resource type represents a local git repository. It is
+the primary resource type used by the git worktree sandbox during plan
+execution.
+
+See [Resource CLI Reference](../reference/resource_cli.md) for all resource
+commands.
+
+---
+
+### `agents project create` and `agents project link-resource`
+
+Create a project and link one or more resources to it.
+
+```bash
+agents project create local/my-project --description "My project"
+agents project link-resource local/my-project local/my-repo
+```
+
+Projects group resources together and provide the context for plan execution.
+Multiple resources can be linked to a single project.
+
+---
+
+### `agents plan use`
+
+Create a plan record from an action template and one or more projects.
+
+```bash
+agents plan use local/my-action local/my-project
+```
+
+This creates a plan in the `strategize/queued` state and persists it to
+SQLite. The plan is assigned a ULID identifier.
+
+See [Plan CLI Reference](../reference/plan_cli.md) for full options including
+`--arg`, `--automation-profile`, and `--invariant` flags.
+
+---
+
+### `agents plan execute <plan_id>`
+
+Run the Strategize and Execute phases of the plan lifecycle. This invokes
+the actor-based LLM path and writes generated file changes to the sandbox.
+
+```bash
+agents plan execute 01HXYZ1234567890ABCDEFGH
+```
+
+The executor:
+1. Runs the **Strategize** phase — the strategy actor produces a decision
+   tree of tasks.
+2. Runs the **Execute** phase — the execution actor generates file changes
+   and writes them to the git worktree sandbox.
+
+When the automation profile permits auto-apply (e.g. `ci` or `full-auto`),
+`plan execute` drives the plan all the way to `applied` in a single call.
+
+---
+
+### `agents plan diff <plan_id>`
+
+Show pending changes in the sandbox as a unified diff.
+
+```bash
+agents plan diff 01HXYZ1234567890ABCDEFGH
+```
+
+Displays the ChangeSet as a unified diff so you can review what the LLM
+generated before committing it to the real repository.
+
+---
+
+### `agents plan apply <plan_id>`
+
+Merge sandbox changes into the real git repository with a commit.
+
+```bash
+agents plan apply 01HXYZ1234567890ABCDEFGH
+```
+
+For git-checkout resources, this uses the **git worktree sandbox** strategy:
+changes are merged via `git merge` from the isolated worktree branch into
+the original branch. Non-git projects fall back to flat file copy.
+
+Pass `--yes` / `-y` to skip the confirmation prompt in CI pipelines.
+
+---
+
+## Git Worktree Sandbox
+
+The git worktree sandbox provides isolated working directories for plan
+execution. Each plan gets its own branch (`cleveragents/plan-<plan_id>`)
+and a temporary worktree directory.
+
+**Lifecycle:**
+
+```
+PENDING → CREATED → ACTIVE → COMMITTED → CLEANED_UP
+                       │
+                       └→ ROLLED_BACK → CLEANED_UP
+```
+
+On `plan apply`, the sandbox branch is merged back into the original branch
+and the worktree is cleaned up. On failure or cancellation, the worktree is
+discarded without touching the original branch.
+
+See [Git Worktree Sandbox](git-worktree-sandbox.md) for the full API
+reference.
+
+---
+
+## Domain Models
+
+All domain models introduced in M1 use **Pydantic v2** with `frozen=True`:
+
+| Model | Description |
+|-------|-------------|
+| `ActionRecord` | Persisted action template |
+| `ActionArgument` | Typed argument definition for an action |
+| `ResourceRecord` | Registered resource instance |
+| `ProjectRecord` | Project grouping resources |
+| `ProjectResourceLink` | Many-to-many link between projects and resources |
+| `PlanRecord` | Plan instance with state machine |
+| `ChangeSet` | Set of file changes produced by execution |
+
+Frozen models ensure that domain objects are never mutated after creation,
+preventing accidental state corruption in multi-phase workflows.
+
+---
+
+## SQLite Persistence and Alembic Migrations
+
+All records are persisted to a local SQLite database (default:
+`~/.config/cleveragents/cleveragents.db`). Schema migrations are managed
+by **Alembic** and run automatically on first use.
+
+The database path can be overridden via the `CLEVERAGENTS_DB_PATH`
+environment variable or the `database.path` config key.
+
+---
+
+## State Machine
+
+Plans follow a strict state machine with the following phases and states:
+
+| Phase | States |
+|-------|--------|
+| `strategize` | `queued` → `in_progress` → `complete` / `errored` |
+| `execute` | `queued` → `in_progress` → `complete` / `errored` |
+| `apply` | `queued` → `in_progress` → `applied` / `errored` |
+
+Terminal states: `applied`, `cancelled`, `errored`.
+
+---
+
+## End-to-End Example
+
+```bash
+# 1. Register an action
+agents action create --config action.yaml
+
+# 2. Register a git repository resource
+agents resource add git-checkout local/my-repo --path /home/user/my-repo
+
+# 3. Create a project and link the resource
+agents project create local/my-project
+agents project link-resource local/my-project local/my-repo
+
+# 4. Create a plan
+agents plan use local/my-action local/my-project
+
+# 5. Execute the plan (Strategize + Execute phases)
+agents plan execute 01HXYZ1234567890ABCDEFGH
+
+# 6. Review the generated changes
+agents plan diff 01HXYZ1234567890ABCDEFGH
+
+# 7. Apply the changes to the real repository
+agents plan apply 01HXYZ1234567890ABCDEFGH
+```
+
+---
+
+## Related
+
+- [Plan CLI Reference](../reference/plan_cli.md)
+- [Action CLI Reference](../reference/action_cli.md)
+- [Resource CLI Reference](../reference/resource_cli.md)
+- [Git Worktree Sandbox](git-worktree-sandbox.md)
+- [ADR-006 Plan Lifecycle](../adr/ADR-006-plan-lifecycle.md)
+- [ADR-008 Resource System](../adr/ADR-008-resource-system.md)
+- [ADR-009 Project Model](../adr/ADR-009-project-model.md)
+- [ADR-015 Sandbox & Checkpoint](../adr/ADR-015-sandbox-and-checkpoint.md)
+- [ADR-019 Storage & Persistence](../adr/ADR-019-storage-and-persistence.md)
@@ -0,0 +1,305 @@
+# v3.1.0 — M2: Actor Compiler + Full LLM Integration
+
+**Milestone:** v3.1.0 — M2 (CLOSED)
+**Released:** 2025-Q4
+
+This milestone made Actor YAML files fully operational. Custom actors compile
+into live LangGraph graphs, the tool router normalizes calls across providers,
+the validation runner enforces resource-attached validations, and the MCP
+adapter connects to external tool servers.
+
+---
+
+## Overview
+
+M2 built on the M1 plan lifecycle by introducing a full actor compilation
+pipeline and LLM integration layer:
+
+```
+actor.yaml (version: "3")
+      │
+      ▼
+ActorLoader (parse + validate via Pydantic)
+      │
+      ▼
+ActorCompiler (GRAPH type → LangGraph StateGraph)
+      │
+      ▼
+LangGraph Runtime
+      │
+      ├── ToolCallRouter (OpenAI / Anthropic / LangChain normalization)
+      │
+      ├── ValidationRunner (required + informational validations)
+      │
+      └── MCPToolAdapter (external tool servers)
+```
+
+---
+
+## Actor YAML Schema (version: "3")
+
+Actor YAML files use `version: "3"` and support three actor types:
+
+| Type | Description |
+|------|-------------|
+| `llm` | Single LLM invocation node |
+| `tool` | Tool execution node |
+| `graph` | Multi-node LangGraph workflow |
+
+**Minimal LLM actor:**
+
+```yaml
+version: "3"
+name: local/my-llm-actor
+type: llm
+description: A simple LLM actor
+model: gpt-4
+```
+
+**Minimal GRAPH actor:**
+
+```yaml
+version: "3"
+name: local/my-pipeline
+type: graph
+description: A multi-node pipeline
+model: gpt-4
+
+route:
+  nodes:
+    - id: planner
+      type: agent
+      name: Planner
+      description: Plans the work
+    - id: executor
+      type: agent
+      name: Executor
+      description: Executes the plan
+  edges:
+    - from_node: planner
+      to_node: executor
+  entry_node: planner
+  exit_nodes:
+    - executor
+```
+
+See [Actor Configuration Reference](../reference/actor_config.md) and
+[Actor YAML Schema](../reference/actors_schema.md) for the full field listing.
+
+---
+
+## `agents actor add --config actor.yaml`
+
+Load and register a custom actor from a YAML file.
+
+```bash
+agents actor add local/my-actor --config ./actors/my-actor.yaml
+```
+
+The loader:
+1. Parses the YAML file.
+2. Validates the schema via Pydantic (all fields, types, and constraints).
+3. For GRAPH actors, validates the route topology (no cycles, reachable nodes,
+   valid entry/exit points).
+4. Persists the actor to the registry.
+
+Use `--update` to replace an existing actor, `--unsafe` to register actors
+marked as unsafe, and `--set-default` to make the actor the default.
+
+See [Actor CLI Reference](../reference/actor_cli.md) for all options.
+
+---
+## Actor Compiler: GRAPH → LangGraph
+
+GRAPH-type actors compile into LangGraph `StateGraph` structures via the
+`ActorCompiler`. The compilation pipeline:
+
+1. **Input validation** — Accepts `ActorConfigSchema` with `type=GRAPH`.
+2. **Reference validation** — All node IDs in edges, entry, and exit points
+   are checked against the declared node set.
+3. **Cycle detection** — Intra-graph and cross-actor subgraph cycles are
+   detected and rejected.
+4. **Node mapping** — Each `NodeDefinition` maps to a LangGraph `NodeConfig`.
+5. **Edge mapping** — Each `EdgeDefinition` maps to a LangGraph `Edge`.
+6. **LSP binding extraction** — Per-node LSP bindings are extracted into
+   `LspBinding` objects.
+7. **Metadata assembly** — Returns a `CompiledActor` with nodes, edges, and
+   `CompilationMetadata`.
+
+**Node type mapping:**
+
+| Actor Node Type | LangGraph NodeType |
+|---|---|
+| `agent` | `AGENT` |
+| `tool` | `TOOL` |
+| `conditional` | `CONDITIONAL` |
+| `subgraph` | `SUBGRAPH` |
+
+See [Actor Compiler Reference](../reference/actor_compiler.md) for the full
+API reference.
+
+---
+
+## Skill Registry and Tool Lifecycle
+
+Skills provide named collections of tools that actors can use. The skill
+registry manages the full lifecycle:
+
+```bash
+# Add a skill
+agents skill add local/my-skill --config skill.yaml
+
+# List registered skills
+agents skill list
+
+# Show skill details
+agents skill show local/my-skill
+
+# Remove a skill
+agents skill remove local/my-skill
+```
+
+Tools within skills are registered in the `ToolRegistry` and made available
+to actors during plan execution. See [Skill CLI Reference](../reference/skill_cli.md)
+for all commands.
+
+---
+## MCP Adapter
+
+The `MCPToolAdapter` connects to external MCP (Model Context Protocol) tool
+servers and registers their tools in the `ToolRegistry`.
+
+**Server configuration:**
+
+```yaml
+name: local/my-mcp-server
+transport: stdio
+command: npx
+args: ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/dir"]
+```
+
+**Lifecycle:**
+
+```python
+adapter = MCPToolAdapter(config)
+adapter.connect()
+tools = adapter.discover_tools()
+adapter.register_tools(registry, namespace="mcp")
+# ... use tools in plan execution ...
+adapter.disconnect()
+```
+
+Supported transports: `stdio` (child process), `sse`, `streamable-http`.
+
+See [MCP Tool Adapter Reference](../reference/mcp_adapter.md) for the full
+API reference.
+
+---
+
+## Tool Call Router
+
+The `ToolCallRouter` normalizes tool calls across LLM providers. It accepts
+OpenAI, Anthropic, and LangChain tool call formats and produces normalized
+`NormalizedToolCallResult` objects.
+
+**Provider format mapping:**
+
+| Provider | Arguments key | Arguments type |
+|----------|---------------|----------------|
+| OpenAI | `arguments` | JSON string |
+| Anthropic | `input` | dict |
+| LangChain | `args` | dict |
+
+The router automatically detects the provider format and normalizes the call
+before dispatching to the `ToolRunner`.
+
+See [Tool Call Router Reference](../reference/tool_router.md) for the full
+API reference.
+
+---
+
+## Validation Runner
+
+The validation runner executes resource-attached validations before the
+apply gate. Validations are classified as:
+
+- **Required** — must pass for `plan apply` to proceed.
+- **Informational** — run and reported but do not block apply.
+
+The `ApplyValidationSummary.all_required_passed` property returns `False`
+when zero validations were run (empty-run guard), ensuring apply is always
+blocked unless at least one validation was executed.
+
+---
+
+## Multi-File Generation and ChangeSet
+
+The execution actor generates a `ChangeSet` containing multiple file
+operations. M2 ensures that multi-file generation produces a correct
+`ChangeSet` with:
+
+- `added_files` — new files created by the actor.
+- `modified_files` — existing files changed by the actor.
+- `deleted_files` — files removed by the actor.
+
+The ChangeSet is written to the git worktree sandbox and can be reviewed
+with `agents plan diff` before applying.
+
+---
+## End-to-End Example with Custom Actor
+
+```bash
+# 1. Write an actor YAML file
+cat > my-actor.yaml << 'EOF'
+version: "3"
+name: local/code-reviewer
+type: graph
+description: Multi-step code review pipeline
+model: gpt-4
+
+route:
+  nodes:
+    - id: analyzer
+      type: agent
+      name: Code Analyzer
+      description: Analyzes code quality
+    - id: reporter
+      type: agent
+      name: Report Writer
+      description: Writes the review report
+  edges:
+    - from_node: analyzer
+      to_node: reporter
+  entry_node: analyzer
+  exit_nodes:
+    - reporter
+EOF
+
+# 2. Register the actor
+agents actor add local/code-reviewer --config my-actor.yaml
+
+# 3. Create an action using the custom actor
+agents action create --config action.yaml  # references local/code-reviewer
+
+# 4. Run a plan using the custom actor
+agents plan use local/review-action local/my-project
+agents plan execute 01HXYZ1234567890ABCDEFGH
+agents plan diff 01HXYZ1234567890ABCDEFGH
+agents plan apply 01HXYZ1234567890ABCDEFGH
+```
+
+---
+## Related
+
+- [Actor CLI Reference](../reference/actor_cli.md)
+- [Actor Configuration Reference](../reference/actor_config.md)
+- [Actor YAML Schema](../reference/actors_schema.md)
+- [Actor Compiler Reference](../reference/actor_compiler.md)
+- [MCP Tool Adapter Reference](../reference/mcp_adapter.md)
+- [Tool Call Router Reference](../reference/tool_router.md)
+- [Skill CLI Reference](../reference/skill_cli.md)
+- [ADR-010 Actor & Agent Architecture](../adr/ADR-010-actor-and-agent-architecture.md)
+- [ADR-011 Tool System](../adr/ADR-011-tool-system.md)
+- [ADR-022 LangChain/LangGraph Integration](../adr/ADR-022-langchain-langgraph-integration.md)
+- [ADR-029 Model Context Protocol (MCP)](../adr/ADR-029-model-context-protocol.md)
+- [ADR-031 Actor Abstraction Definition](../adr/ADR-031-actor-abstraction-definition.md)
@@ -30,6 +30,8 @@ nav:
    - Invariant Reconciliation: modules/invariant-reconciliation.md
    - ACMS Context Hydration: modules/context-hydration.md
    - Git Worktree Sandbox: modules/git-worktree-sandbox.md
+    - M1: Minimal Local Workflow: modules/milestone-v3.0.0-local-workflow.md
+    - M2: Actor Compiler + LLM Integration: modules/milestone-v3.1.0-actor-compiler.md
  - Development:
    - Agent System Specification: development/agent-system-specification.md
    - CI/CD Pipeline: development/ci-cd.md