Beta → 1.0 hardening + data-loss repair (24 hardening items + C1-C6)

CLAUDE.md

@@ -122,15 +122,8 @@ loci hook install                            # Register hooks to ~/.claude/setti
 <!-- BEGIN CODEATRIUM -->
 ## Past Memory Search (codeatrium)
 
-IMPORTANT: Command usage is injected automatically at session start via `loci prime` (SessionStart hook).
+IMPORTANT: Full usage instructions are injected automatically at session start via `loci prime` (SessionStart hook).
 If not in context, run `loci prime`.
-
-### Rules
-
-1. **Search before implementing** — always check if something was discussed or built before starting work.
-2. **Check symbols when you lack context** — run `loci context --symbol` before changing a function you don't have enough background on.
-3. **Use technical terms** — queries with exact symbol names, error messages, or parameter names yield better results.
-4. **Follow up with `loci show`** — when `exchange_core` is ambiguous, fetch the full verbatim conversation.
 <!-- END CODEATRIUM -->
 
 ---

README.md

@@ -87,11 +87,12 @@ Agent instructions are injected automatically — no manual setup required:
 | `loci index` | Index new session logs |
 | `loci distill [--limit N]` | Distill undistilled exchanges via LLM |
 | `loci search "query" --json` | Semantic search (agent-facing) |
-| `loci context --symbol "name" --json` | Code symbol → past conversations |
+| `loci context --symbol "name" --json` | Code symbol → past conversations (lightweight; add `--full` for verbatim text) |
 | `loci show "<ref>" --json` | Retrieve verbatim conversation |
 | `loci status` | Show index state |
 | `loci server start/stop/status` | Embedding server management |
 | `loci hook install` | Re-register hooks (normally already done by `loci init`) |
+| `loci hook uninstall` | Remove codeatrium hooks from `settings.json` |
 
 ## Automation (Claude Code Hooks)
 

src/codeatrium/cli/__init__.py

@@ -167,7 +167,7 @@ def init(
                 try:
                     con.execute(
                         f"""
-                        UPDATE exchanges SET distilled_at = 'skipped'
+                        UPDATE exchanges SET distilled_at = 'skipped', distill_status = 'skipped'
                         WHERE distilled_at IS NULL
                         AND id IN (
                             SELECT id FROM exchanges
@@ -237,14 +237,16 @@ def _on_progress(
                 else:
                     typer.echo(f"  [{cur}/{tot}] distilled", err=True)
 
-            count = distill_all(
+            count, err_count = distill_all(
                 db,
                 model=cfg.distill_model,
                 on_progress=_on_progress,
                 project_root=str(root),
                 distill_min_chars=cfg.distill_min_chars,
             )
             typer.echo(f"Distilled {count} exchange(s).")
+            if err_count > 0:
+                typer.echo(f"{err_count} exchange(s) failed — see errors above.", err=True)
         except KeyboardInterrupt:
             typer.echo(
                 "\n⚠ Distillation interrupted. "

src/codeatrium/cli/distill_cmd.py

@@ -14,6 +14,7 @@ def distill(
     ] = None,
 ) -> None:
     """未蒸留の exchange を claude -p で蒸留して palace_objects を生成する"""
+    import fcntl
     import os
 
     from codeatrium.config import load_config
@@ -30,31 +31,14 @@ def distill(
 
     lock_path = db.parent / "distill.lock"
 
-    # ロック取得: O_CREAT | O_EXCL で原子的に作成（TOCTOU 防止）
+    # ロック取得: fcntl.flock で排他ロック（LOCK_NB: 非ブロッキング）
+    fd = os.open(str(lock_path), os.O_CREAT | os.O_RDWR, 0o600)
     try:
-        fd = os.open(lock_path, os.O_CREAT | os.O_EXCL | os.O_WRONLY)
-        os.write(fd, str(os.getpid()).encode())
+        fcntl.flock(fd, fcntl.LOCK_EX | fcntl.LOCK_NB)
+    except BlockingIOError:
         os.close(fd)
-    except FileExistsError:
-        # 既存ロックのプロセスが生きているか確認
-        try:
-            existing_pid = int(lock_path.read_text().strip())
-            os.kill(existing_pid, 0)
-            typer.echo(
-                f"loci distill is already running (PID {existing_pid}). Exiting.",
-                err=True,
-            )
-            raise typer.Exit(0)
-        except (ValueError, ProcessLookupError, PermissionError):
-            # stale lock — 再取得
-            lock_path.unlink(missing_ok=True)
-            try:
-                fd = os.open(lock_path, os.O_CREAT | os.O_EXCL | os.O_WRONLY)
-                os.write(fd, str(os.getpid()).encode())
-                os.close(fd)
-            except FileExistsError:
-                typer.echo("loci distill: lost lock race after stale cleanup. Exiting.", err=True)
-                raise typer.Exit(0)
+        typer.echo("loci distill is already running. Exiting.", err=True)
+        raise typer.Exit(0)
 
     def _on_progress(cur: int, tot: int, error: str | None = None) -> None:
         if error:
@@ -63,7 +47,16 @@ def _on_progress(cur: int, tot: int, error: str | None = None) -> None:
             typer.echo(f"  [{cur}/{tot}] distilled", err=True)
 
     try:
-        count = distill_all(
+        from codeatrium.db import check_drift
+
+        drifts = check_drift(db)
+        for key, recorded, current in drifts:
+            typer.echo(
+                f"[warn] {key} changed ({recorded} -> {current}). Re-index recommended.",
+                err=True,
+            )
+
+        count, err_count = distill_all(
             db,
             limit=limit,
             model=cfg.distill_model,
@@ -72,5 +65,8 @@ def _on_progress(cur: int, tot: int, error: str | None = None) -> None:
             distill_min_chars=cfg.distill_min_chars,
         )
         typer.echo(f"Distilled {count} exchange(s).")
+        if err_count > 0:
+            typer.echo(f"{err_count} exchange(s) failed — see errors above.", err=True)
     finally:
-        lock_path.unlink(missing_ok=True)
+        fcntl.flock(fd, fcntl.LOCK_UN)
+        os.close(fd)

src/codeatrium/cli/hook_cmd.py

@@ -22,3 +22,12 @@ def hook_install() -> None:
     cfg = load_config(find_project_root())
     _changed, message = install_hooks(batch_limit=cfg.distill_batch_limit)
     typer.echo(message)
+
+
+@hook_app.command("uninstall")
+def hook_uninstall() -> None:
+    """Claude Code の settings.json から codeatrium 関連フックをすべて除去する。"""
+    from codeatrium.hooks import uninstall_hooks
+
+    _changed, message = uninstall_hooks()
+    typer.echo(message)

src/codeatrium/cli/index_cmd.py

@@ -32,6 +32,10 @@ def index(
         raise typer.Exit(1)
 
     init_db(db)
+    from codeatrium.db import check_drift
+    drifts = check_drift(db)
+    for key, recorded, current in drifts:
+        typer.echo(f"[warn] {key} changed ({recorded} -> {current}). Re-index recommended.", err=True)
     cfg = load_config(root)
 
     target_dir = path or resolve_claude_projects_path(root)

src/codeatrium/cli/prime_cmd.py

@@ -12,44 +12,40 @@
 PRIME_TEXT = """\
 ## Past Memory Search (codeatrium)
 
-Use `loci search` to find past implementations, decisions, and code locations.
+codeatrium records every past conversation turn, decision, and code location. Retrieve that memory **before** acting — not after.
 
-### When to use
+### When to act (agent-initiated triggers)
 
-- When asked "where did we implement X?" or "where is X?"
-- When checking if a similar bug was fixed before
-- When verifying if a feature already exists
-- When looking up the reasoning behind a past design decision
-- Before editing code you lack context about — use `loci context --symbol` to review past discussions
-- Before refactoring or changing the behavior of a function — use `loci context --symbol` to check past design decisions
+- **Before editing or refactoring a function** — recall past design decisions and known constraints for that symbol.
+- **Before starting a new implementation** — check if similar work was done before; reuse decisions and avoid re-debating settled choices.
+- **When you encounter a known or recurring error** — search for past fixes; the solution may already be documented.
 
-### Commands
+### Search — semantic query over past conversations
 
 ```bash
-# Semantic search
-loci search "query" --json --limit 5
+# Find past discussions, decisions, or implementations
+loci search "BM25 RRF fusion ranking" --json --limit 5
 
-# Reverse lookup: code symbol -> past conversations
-loci context --symbol "symbol_name" --json
-
-# Retrieve verbatim conversation (use verbatim_ref from search results)
+# Retrieve verbatim exchange (use verbatim_ref from search results)
 loci show "<verbatim_ref>" --json
+```
+
+### Context — reverse lookup from code symbol to past conversations
+
+Touching a symbol = recalling memory about that symbol. Before changing any function or class, look up what was decided about it.
+
+```bash
+# Retrieve all past conversations that involved this symbol
+loci context --symbol "SymbolResolver.extract" --json
 ```\
 """
 
 CLAUDE_MD_SECTION = f"""\
 {BEGIN_MARKER}
 ## Past Memory Search (codeatrium)
 
-IMPORTANT: Command usage is injected automatically at session start via `loci prime` (SessionStart hook).
+IMPORTANT: Full usage instructions are injected automatically at session start via `loci prime` (SessionStart hook).
 If not in context, run `loci prime`.
-
-### Rules
-
-1. **Search before implementing** — always check if something was discussed or built before starting work.
-2. **Check symbols when you lack context** — run `loci context --symbol` before changing a function you don't have enough background on.
-3. **Use technical terms** — queries with exact symbol names, error messages, or parameter names yield better results.
-4. **Follow up with `loci show`** — when `exchange_core` is ambiguous, fetch the full verbatim conversation.
 {END_MARKER}\
 """
 

src/codeatrium/cli/search_cmd.py

@@ -25,6 +25,11 @@ def search(
         typer.echo("Not initialized. Run `loci init` first.", err=True)
         raise typer.Exit(1)
 
+    from codeatrium.db import check_drift
+    drifts = check_drift(db)
+    for key, recorded, current in drifts:
+        typer.echo(f"[warn] {key} changed ({recorded} -> {current}). Re-index recommended.", err=True)
+
     embedder = Embedder()
     query_vec = embedder.embed(query)
     results = search_combined(db, query, query_vec, limit=limit)
@@ -62,6 +67,7 @@ def context(
     ],
     limit: Annotated[int, typer.Option("--limit", "-n", help="返す件数")] = 5,
     json_output: Annotated[bool, typer.Option("--json", help="JSON で出力")] = False,
+    full: Annotated[bool, typer.Option("--full", help="全文（user_content / agent_content）を含める")] = False,
 ) -> None:
     """シンボル名から関連する過去会話を逆引きする"""
     from codeatrium.db import get_connection
@@ -87,10 +93,13 @@ def context(
             e.user_content,
             e.agent_content,
             p.exchange_core,
-            p.specific_context
+            p.specific_context,
+            c.source_path,
+            e.ply_start
         FROM symbols s
         JOIN palace_objects p ON p.id = s.palace_object_id
         JOIN exchanges e ON e.id = p.exchange_id
+        JOIN conversations c ON c.id = e.conversation_id
         WHERE s.symbol_name LIKE ?
         LIMIT ?
         """,
@@ -103,8 +112,9 @@ def context(
         return
 
     if json_output:
-        output = [
-            {
+        output = []
+        for r in rows:
+            base = {
                 "symbol_name": r["symbol_name"],
                 "symbol_kind": r["symbol_kind"],
                 "file_path": r["file_path"],
@@ -113,11 +123,12 @@ def context(
                 "exchange_id": r["exchange_id"],
                 "exchange_core": r["exchange_core"],
                 "specific_context": r["specific_context"],
-                "user_content": r["user_content"],
-                "agent_content": r["agent_content"],
+                "verbatim_ref": f"{r['source_path']}:ply={r['ply_start']}",
             }
-            for r in rows
-        ]
+            if full:
+                base["user_content"] = r["user_content"]
+                base["agent_content"] = r["agent_content"]
+            output.append(base)
         typer.echo(json.dumps(output, ensure_ascii=False, indent=2))
     else:
         for i, r in enumerate(rows, 1):
@@ -126,3 +137,4 @@ def context(
             typer.echo(f"    {r['signature']}")
             if r["exchange_core"]:
                 typer.echo(f"    Core: {r['exchange_core']}")
+            typer.echo(f"    {r['source_path']}:ply={r['ply_start']}")

src/codeatrium/cli/server_cmd.py

@@ -6,11 +6,14 @@
 
 server_app = typer.Typer(help="embedding サーバー管理")
 
+_SERVER_STARTUP_POLL_ATTEMPTS: int = 150  # サーバー起動確認のポーリング回数（0.2秒 × 150 = 最大30秒待機）
+
 
 @server_app.command("start")
 def server_start() -> None:
     """embedding サーバーをバックグラウンドで起動する"""
     import json as _json
+    import os
     import socket as _socket
     import subprocess
 
@@ -36,8 +39,20 @@ def server_start() -> None:
                     return
         except Exception:
             sock.unlink(missing_ok=True)
+            server_pid_path(root).unlink(missing_ok=True)
 
     pid_path = server_pid_path(root)
+    if pid_path.exists():
+        try:
+            _pid = int(pid_path.read_text().strip())
+            os.kill(_pid, 0)
+        except (ProcessLookupError, ValueError):
+            # プロセス不在 or 不正な PID → stale とみなして除去
+            pid_path.unlink(missing_ok=True)
+        except PermissionError:
+            # 別ユーザーのプロセスが生存 → 触らない
+            pass
+
     proc = subprocess.Popen(
         [_loci_python(), "-m", "codeatrium.embedder_server", str(sock)],
         stdout=subprocess.DEVNULL,
@@ -48,7 +63,7 @@ def server_start() -> None:
 
     import time
 
-    for i in range(150):
+    for i in range(_SERVER_STARTUP_POLL_ATTEMPTS):
         if sock.exists():
             typer.echo(f"Server started (PID {proc.pid})")
             return

src/codeatrium/cli/status_cmd.py

@@ -25,12 +25,26 @@ def status(
     con = get_connection(db)
     total = con.execute("SELECT COUNT(*) FROM exchanges").fetchone()[0]
     distilled = con.execute(
-        "SELECT COUNT(*) FROM exchanges WHERE distilled_at IS NOT NULL"
+        "SELECT COUNT(*) FROM exchanges WHERE distill_status = 'distilled'"
+    ).fetchone()[0]
+    skipped = con.execute(
+        "SELECT COUNT(*) FROM exchanges WHERE distill_status = 'skipped'"
+    ).fetchone()[0]
+    pending = con.execute(
+        "SELECT COUNT(*) FROM exchanges WHERE distill_status = 'pending'"
     ).fetchone()[0]
     palace_count = con.execute("SELECT COUNT(*) FROM palace_objects").fetchone()[0]
     symbol_count = con.execute("SELECT COUNT(*) FROM symbols").fetchone()[0]
     con.close()
 
+    from codeatrium.db import check_drift
+    drifts = check_drift(db)
+    for key, recorded, current in drifts:
+        typer.echo(
+            f"[drift] {key}: recorded={recorded}, current={current} — re-index recommended",
+            err=True,
+        )
+
     db_size_bytes = db.stat().st_size
     db_size_kb = db_size_bytes / 1024
 
@@ -41,7 +55,8 @@ def status(
                     "db_path": str(db),
                     "exchanges": total,
                     "distilled": distilled,
-                    "undistilled": total - distilled,
+                    "skipped": skipped,
+                    "pending": pending,
                     "palace_objects": palace_count,
                     "symbols": symbol_count,
                     "db_size_kb": round(db_size_kb, 1),
@@ -53,7 +68,7 @@ def status(
     else:
         typer.echo(f"DB: {db} ({db_size_kb:.1f} KB)")
         typer.echo(
-            f"Exchanges : {total} total, {distilled} distilled, {total - distilled} pending"
+            f"Exchanges : {total} total | {distilled} distilled, {skipped} skipped, {pending} pending"
         )
         typer.echo(f"Palace    : {palace_count}")
         typer.echo(f"Symbols   : {symbol_count}")