diff --git a/docs/local_agent_loop.md b/docs/local_agent_loop.md
index ed9fb54..8b8f7f3 100644
--- a/docs/local_agent_loop.md
+++ b/docs/local_agent_loop.md
@@ -14,22 +14,28 @@ The default coder is Claude and the default reviewer is Codex. Reverse the direc
 ## Architecture
 
 The tool is a local orchestrator. It does not call model APIs directly; it shells
-out to locally authenticated agent and GitHub CLIs from separate checkouts.
+out to locally authenticated agent and GitHub CLIs from separate checkouts. The
+only durable state it creates is local: agent logs in the active coder checkout
+and optional advisory memory in a repo-scoped cache directory.
 
 ```mermaid
 flowchart LR
     User[Developer terminal] --> CLI[agent-loop CLI<br/>cli.py]
-    CLI --> Config[Config and workdir setup<br/>config.py / workdirs.py]
-    CLI --> Orchestrator[Issue, task, and PR loops<br/>orchestrator.py]
+    CLI --> Config[Config validation<br/>config.py]
+    Config --> Orchestrator[Issue, task, and PR loops<br/>orchestrator.py]
 
-    Config --> Workdirs[(Agent checkouts<br/>Claude / Codex / Gemini)]
-    Config --> Logs[(.agent-loop-logs)]
+    Orchestrator --> Workdirs[Workdir setup and validation<br/>workdirs.py / git / gh repo clone]
+    Orchestrator --> Memory[Agent memory preparation<br/>memory.py]
+
+    Workdirs --> AgentDirs[(Separate agent checkouts<br/>Claude / Codex / Gemini)]
+    Memory --> MemoryCache[(Repo-scoped memory cache<br/>summary / architecture / tests)]
 
     Orchestrator --> Prompts[Prompt builders<br/>prompts.py]
-    Orchestrator --> Protocol[Marker parsing<br/>protocol.py]
+    Orchestrator --> Protocol[Marker and follow-up parsing<br/>protocol.py]
     Orchestrator --> Registry[Agent registry<br/>agents/registry.py]
     Orchestrator --> GitHubOps[GitHub operations<br/>github.py]
-    Orchestrator --> OptionalTests[Optional local test command]
+    Orchestrator --> OptionalTests[Optional local test command<br/>--test-command]
+    Orchestrator --> Followups[Approved follow-up handling<br/>summaries / issues / same-PR fixes]
 
     Registry --> Claude[Claude backend<br/>claude]
     Registry --> Codex[Codex backend<br/>codex exec]
@@ -44,10 +50,11 @@ flowchart LR
     Runner --> AgentCLIs[Local agent CLIs]
     Runner --> GhCLI[gh CLI]
     Runner --> TestCmd[Local test process]
+    Runner --> Logs
 
-    AgentCLIs --> Workdirs
-    AgentCLIs --> Logs
+    AgentCLIs --> AgentDirs
     GhCLI --> GitHub[(GitHub repo<br/>issues / PRs / comments / checks)]
+    Followups --> GitHubOps
 ```
 
 At runtime, the orchestrator drives one of three entrypoints:
@@ -57,12 +64,15 @@ sequenceDiagram
     participant User as Developer
     participant CLI as agent-loop CLI
     participant Orch as Orchestrator
+    participant Memory as Agent memory
     participant Coder as Coder agent CLI
     participant Reviewer as Reviewer agent CLI(s)
     participant GH as GitHub via gh
 
     User->>CLI: agent-loop issue | task | pr
-    CLI->>Orch: validated config and workdirs
+    CLI->>Orch: validated config
+    Orch->>Orch: ensure active agent workdirs
+    Orch->>Memory: prepare advisory repo memory
     alt issue or task
         Orch->>Coder: create or update PR
         Coder-->>Orch: output with AGENT_PR marker
@@ -79,7 +89,14 @@ sequenceDiagram
             Orch->>Coder: address combined feedback
             Coder-->>Orch: AGENT_STATE blocking
             Orch->>GH: post coder update
+        else approved review has same-PR follow-ups in a fix-and mode
+            Orch->>Coder: address same-PR follow-ups
+            Coder-->>Orch: AGENT_STATE blocking
+            Orch->>GH: post coder update
         else all approved
+            opt future follow-ups requested
+                Orch->>GH: summarize follow-ups or create issues
+            end
             Orch->>Orch: run optional local tests
             opt auto-merge enabled
                 Orch->>GH: wait for configured check and merge
diff --git a/src/coding_review_agent_loop/cli.py b/src/coding_review_agent_loop/cli.py
index 383708e..a6c1418 100644
--- a/src/coding_review_agent_loop/cli.py
+++ b/src/coding_review_agent_loop/cli.py
@@ -192,7 +192,7 @@ def add_common(subparser: argparse.ArgumentParser) -> None:
             "--no-agent-memory",
             dest="agent_memory",
             action="store_false",
-            help="Disable repo-local advisory agent memory.",
+            help="Disable repo-scoped advisory agent memory.",
         )
         subparser.add_argument(
             "--refresh-agent-memory",
diff --git a/src/coding_review_agent_loop/memory.py b/src/coding_review_agent_loop/memory.py
index 9c237ee..10464df 100644
--- a/src/coding_review_agent_loop/memory.py
+++ b/src/coding_review_agent_loop/memory.py
@@ -1,4 +1,4 @@
-"""Repo-local advisory memory for repeated agent loop runs."""
+"""Repo-scoped advisory memory for repeated agent loop runs."""
 
 from __future__ import annotations
 
@@ -232,10 +232,44 @@ def _write_architecture_map(path: Path, tracked_files: list[str]) -> None:
     for file in tracked_files:
         head = file.split("/", 1)[0]
         buckets.setdefault(head, []).append(file)
-    lines = ["# Architecture Map", ""]
+    lines = [
+        "# Architecture Map",
+        "",
+        "## Top-level Layout",
+        "",
+    ]
     for head in sorted(buckets):
         examples = ", ".join(buckets[head][:8])
         lines.append(f"- `{head}`: {len(buckets[head])} tracked file(s); examples: {examples}")
+
+    python_files = [
+        file for file in tracked_files if file.startswith("src/") and file.endswith(".py")
+    ]
+    if python_files:
+        lines.extend(["", "## Python Modules", ""])
+        module_buckets: dict[str, list[str]] = {}
+        for file in python_files:
+            parts = file.split("/")
+            if len(parts) >= 4:
+                module = "/".join(parts[:3])
+            else:
+                module = "/".join(parts[:-1]) or "src"
+            module_buckets.setdefault(module, []).append(file)
+        for module in sorted(module_buckets):
+            examples = ", ".join(module_buckets[module][:8])
+            lines.append(f"- `{module}`: {len(module_buckets[module])} file(s); examples: {examples}")
+
+    docs = [file for file in tracked_files if file == "README.md" or file.startswith("docs/")]
+    tests = [file for file in tracked_files if file.startswith("tests/")]
+    workflows = [file for file in tracked_files if file.startswith(".github/workflows/")]
+    if docs or tests or workflows:
+        lines.extend(["", "## Supporting Surfaces", ""])
+        if docs:
+            lines.append(f"- Docs: {', '.join(docs[:8])}")
+        if tests:
+            lines.append(f"- Tests: {', '.join(tests[:8])}")
+        if workflows:
+            lines.append(f"- GitHub Actions: {', '.join(workflows[:8])}")
     path.write_text("\n".join(lines) + "\n", encoding="utf-8")
 
 
diff --git a/tests/test_agent_loop.py b/tests/test_agent_loop.py
index f2dd060..89567f1 100644
--- a/tests/test_agent_loop.py
+++ b/tests/test_agent_loop.py
@@ -704,6 +704,11 @@ def test_agent_memory_is_created_and_added_to_review_prompt(tmp_path):
     assert (memory_dir / "test-profile.md").exists()
     assert (memory_dir / "toolchain.json").exists()
     assert (memory_dir / "last-analyzed-commit").read_text(encoding="utf-8") == "abc123\n"
+    architecture_map = (memory_dir / "architecture-map.md").read_text(encoding="utf-8")
+    assert "## Top-level Layout" in architecture_map
+    assert "## Python Modules" in architecture_map
+    assert "## Supporting Surfaces" in architecture_map
+    assert "`src/coding_review_agent_loop`" in architecture_map
 
     prompt = next(cmd[-1] for cmd, _cwd in runner.commands if cmd[:2] == ["codex", "exec"])
     assert "Agent memory context:" in prompt