spooling 0.2.8__tar.gz → 0.2.9__tar.gz

{spooling-0.2.8/spooling.egg-info → spooling-0.2.9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: spooling
-Version: 0.2.8
+Version: 0.2.9
 Summary: Local session tracker and semantic search for AI coding assistants
 Author: Parsed Analytics, Inc.
 License: MIT
@@ -38,7 +38,7 @@ Dynamic: license-file
 
 Local session tracker and semantic search for AI coding assistants.
 
-Track your AI coding sessions across **OpenAI Codex CLI**, **GitHub Copilot**, **Cursor**, **Windsurf**, **Kiro**, **Google Antigravity**, and **opencode**, all in one place. Get usage stats, cost estimates, per-provider breakdowns, semantic search, and a built-in AI chat agent to explore your history.
+Track your AI coding sessions across **Claude Code**, **OpenAI Codex CLI**, **GitHub Copilot**, **Cursor**, **Windsurf**, **Kiro**, **Google Antigravity**, and **opencode**, all in one place. Get usage stats, cost estimates, per-provider breakdowns, semantic search, and a built-in AI chat agent to explore your history.
 
 **Website:** [spooling.ai](https://spooling.ai)
 
@@ -142,6 +142,7 @@ spooling sync                      # Full sync with embeddings
 spooling sync --no-embed           # Skip embeddings (faster initial sync)
 spooling sync -p cursor            # Sync only Cursor sessions
 spooling sync -p codex             # Sync only Codex CLI sessions
+spooling sync -p claude-code       # Sync only Claude Code sessions
 ```
 
 ### `spooling stats`
@@ -639,6 +640,7 @@ All optional — defaults work out of the box.
 
 | Provider | Data Location | Format |
 |----------|--------------|--------|
+| **Claude Code** | `~/.claude/projects/` (or `$CLAUDE_CONFIG_DIR/projects`) | `<project>/<session-uuid>.jsonl` transcripts with per-message token usage, prompt-cache reads/writes, tool calls and git context |
 | **JSONL Sessions** | `~/.sessions/projects/` | UUID-named JSONL files with conversation history, tool calls, git context |
 | **OpenAI Codex CLI** | `~/.codex/sessions/` | `rollout-*.jsonl` files organized by date |
 | **GitHub Copilot** | `~/Library/Application Support/Code/User/workspaceStorage/` | Chat session JSON from VS Code |
@@ -650,5 +652,7 @@ All optional — defaults work out of the box.
 
 Run `spooling init` to see which providers are detected on your system.
 
+**Claude Code notes.** Sessions are detected automatically; there is nothing to connect. Cost is computed from each message's real token usage (including prompt-cache reads and writes) at the model's list price, so it reflects API-equivalent spend, not what a subscription plan charges. The project name is the session's working directory. To read transcripts from somewhere else, set `CLAUDE_CONFIG_DIR` or add `{"claude-code": "/path/to/projects"}` to `~/.config/spooling/paths.json`. Subagent transcripts (stored in separate files under `<session>/subagents/` by recent Claude Code versions) are not included yet.
+
 Nothing leaves your machine unless you opt in: Spooling Cloud (`spooling push`), a tunnel (`spooling tunnel`), or the Anthropic chat provider. `spooling pricing refresh` downloads public model prices but sends none of your data.
 

{spooling-0.2.8 → spooling-0.2.9}/README.md

@@ -2,7 +2,7 @@
 
 Local session tracker and semantic search for AI coding assistants.
 
-Track your AI coding sessions across **OpenAI Codex CLI**, **GitHub Copilot**, **Cursor**, **Windsurf**, **Kiro**, **Google Antigravity**, and **opencode**, all in one place. Get usage stats, cost estimates, per-provider breakdowns, semantic search, and a built-in AI chat agent to explore your history.
+Track your AI coding sessions across **Claude Code**, **OpenAI Codex CLI**, **GitHub Copilot**, **Cursor**, **Windsurf**, **Kiro**, **Google Antigravity**, and **opencode**, all in one place. Get usage stats, cost estimates, per-provider breakdowns, semantic search, and a built-in AI chat agent to explore your history.
 
 **Website:** [spooling.ai](https://spooling.ai)
 
@@ -106,6 +106,7 @@ spooling sync                      # Full sync with embeddings
 spooling sync --no-embed           # Skip embeddings (faster initial sync)
 spooling sync -p cursor            # Sync only Cursor sessions
 spooling sync -p codex             # Sync only Codex CLI sessions
+spooling sync -p claude-code       # Sync only Claude Code sessions
 ```
 
 ### `spooling stats`
@@ -603,6 +604,7 @@ All optional — defaults work out of the box.
 
 | Provider | Data Location | Format |
 |----------|--------------|--------|
+| **Claude Code** | `~/.claude/projects/` (or `$CLAUDE_CONFIG_DIR/projects`) | `<project>/<session-uuid>.jsonl` transcripts with per-message token usage, prompt-cache reads/writes, tool calls and git context |
 | **JSONL Sessions** | `~/.sessions/projects/` | UUID-named JSONL files with conversation history, tool calls, git context |
 | **OpenAI Codex CLI** | `~/.codex/sessions/` | `rollout-*.jsonl` files organized by date |
 | **GitHub Copilot** | `~/Library/Application Support/Code/User/workspaceStorage/` | Chat session JSON from VS Code |
@@ -614,5 +616,7 @@ All optional — defaults work out of the box.
 
 Run `spooling init` to see which providers are detected on your system.
 
+**Claude Code notes.** Sessions are detected automatically; there is nothing to connect. Cost is computed from each message's real token usage (including prompt-cache reads and writes) at the model's list price, so it reflects API-equivalent spend, not what a subscription plan charges. The project name is the session's working directory. To read transcripts from somewhere else, set `CLAUDE_CONFIG_DIR` or add `{"claude-code": "/path/to/projects"}` to `~/.config/spooling/paths.json`. Subagent transcripts (stored in separate files under `<session>/subagents/` by recent Claude Code versions) are not included yet.
+
 Nothing leaves your machine unless you opt in: Spooling Cloud (`spooling push`), a tunnel (`spooling tunnel`), or the Anthropic chat provider. `spooling pricing refresh` downloads public model prices but sends none of your data.
 

{spooling-0.2.8 → spooling-0.2.9}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "spooling"
-version = "0.2.8"
+version = "0.2.9"
 description = "Local session tracker and semantic search for AI coding assistants"
 readme = "README.md"
 authors = [{ name = "Parsed Analytics, Inc." }]

{spooling-0.2.8 → spooling-0.2.9}/spooling/cli.py

@@ -64,7 +64,7 @@ def init():
 
 @cli.command()
 @click.option("--no-embed", is_flag=True, help="Skip embedding (faster sync)")
-@click.option("--provider", "-p", default=None, help="Only sync a specific provider (jsonl-session, codex, cursor, copilot, windsurf)")
+@click.option("--provider", "-p", default=None, help="Only sync a specific provider (claude-code, jsonl-session, codex, cursor, copilot, windsurf)")
 def sync(no_embed, provider):
     """Sync AI coding sessions to the database."""
     from spooling.ingest import sync as do_sync

{spooling-0.2.8 → spooling-0.2.9}/spooling/mcp_server.py

@@ -161,7 +161,7 @@ def list_traces(
 
     Args:
         limit: Max rows to return (default 25, capped at 200).
-        provider: Filter to one provider id (jsonl-session, codex, cursor, copilot, windsurf, kiro, antigravity, gemini, opencode).
+        provider: Filter to one provider id (claude-code, jsonl-session, codex, cursor, copilot, windsurf, kiro, antigravity, gemini, opencode).
         project: Filter to sessions whose project name matches exactly.
     """
     limit = max(1, min(limit, 200))

{spooling-0.2.8 → spooling-0.2.9}/spooling/parser.py

@@ -304,8 +304,19 @@ def _tool_result_text(block: dict) -> tuple[str, bool | None]:
     return "", is_error
 
 
-def parse_session_file(file_path: Path) -> ParsedSession | None:
-    """Parse a single session JSONL file into messages + trace."""
+def parse_session_file(
+    file_path: Path,
+    provider_id: str = "jsonl-session",
+    project_from_cwd: bool = False,
+) -> ParsedSession | None:
+    """Parse a single session JSONL file into messages + trace.
+
+    ``provider_id`` is stamped on both the session and its trace (the generic
+    ``jsonl-session`` reader and the ``claude-code`` provider share this parser).
+    With ``project_from_cwd`` the project is the working directory's basename,
+    matching Codex/Gemini/opencode, instead of the on-disk folder name (which for
+    Claude Code is the whole path with slashes replaced by dashes).
+    """
     session_id = file_path.stem
     project = file_path.parent.name
 
@@ -344,9 +355,14 @@ def parse_session_file(file_path: Path) -> ParsedSession | None:
         agent_version = agent_version or r.get("version")
         if not model:
             m = (r.get("message") or {}).get("model")
-            if m:
+            # "<synthetic>" marks locally generated placeholder messages
+            # (interrupts, API errors), not a real model.
+            if m and m != "<synthetic>":
                 model = m
 
+    if project_from_cwd and cwd:
+        project = Path(cwd.rstrip("/")).name or project
+
     # --- Build ParsedMessage list (legacy path, with tool details) -------
     messages: list[ParsedMessage] = []
     pending_tool_details: list[ToolCallDetail] = []
@@ -422,6 +438,7 @@ def parse_session_file(file_path: Path) -> ParsedSession | None:
         git_branch=git_branch,
         model=model,
         title=title,
+        provider_id=provider_id,
     )
 
     return ParsedSession(
@@ -435,6 +452,7 @@ def parse_session_file(file_path: Path) -> ParsedSession | None:
         agent_version=agent_version,
         model=model,
         title=title,
+        provider_id=provider_id,
         trace=trace,
     )
 
@@ -467,6 +485,7 @@ def _build_trace(
     git_branch: str | None,
     model: str | None,
     title: str | None,
+    provider_id: str = "jsonl-session",
 ) -> Trace:
     """Build a Trace with span tree from session records.
 
@@ -479,7 +498,7 @@ def _build_trace(
                 └─ tool ...
     """
     tb = TraceBuilder(
-        provider_id="jsonl-session",
+        provider_id=provider_id,
         session_id=session_id,
         project=project,
         cwd=cwd,

{spooling-0.2.8 → spooling-0.2.9}/spooling/pricing.py

@@ -48,6 +48,7 @@ _GEMINI_ZERO_MINOR = re.compile(r"^(gemini-\d+)-0(-)")
 # "auto"), substitute the provider's documented default before pricing.
 # Source: each provider's docs, picked as the most-common-actual-model.
 PROVIDER_DEFAULT_MODEL: dict[str, str] = {
+    "claude-code":   "claude-sonnet-4-5",
     "copilot":       "gpt-4o",
     "cursor":        "gpt-4o",
     "windsurf":      "gpt-4o",

{spooling-0.2.8 → spooling-0.2.9}/spooling/providers/__init__.py

@@ -2,6 +2,7 @@
 
 from spooling.providers.base import Provider, RemoteProvider, PROVIDER_REGISTRY
 from spooling.providers.session_file import SessionFileProvider
+from spooling.providers.claude_code import ClaudeCodeProvider
 from spooling.providers.codex import CodexProvider
 from spooling.providers.cursor import CursorProvider
 from spooling.providers.copilot import CopilotProvider
@@ -19,6 +20,7 @@ __all__ = [
     "RemoteProvider",
     "PROVIDER_REGISTRY",
     "SessionFileProvider",
+    "ClaudeCodeProvider",
     "CodexProvider",
     "CursorProvider",
     "CopilotProvider",

spooling-0.2.9/spooling/providers/claude_code.py

@@ -0,0 +1,65 @@
+"""Claude Code provider: reads session transcripts from ``~/.claude/projects``.
+
+Claude Code writes one JSONL transcript per session at
+``<config dir>/projects/<encoded-cwd>/<session-uuid>.jsonl``, where the encoded
+cwd is the working directory with ``/`` replaced by ``-``. Each line is a
+``user`` / ``assistant`` record (plus bookkeeping records the parser ignores)
+carrying per-message token ``usage`` including prompt-cache reads and writes,
+which is what makes cost tracking accurate.
+
+The config directory is ``$CLAUDE_CONFIG_DIR`` when set, otherwise ``~/.claude``
+(falling back to ``~/.config/claude`` where newer installs put it). A different
+location can be pinned in ``~/.config/spooling/paths.json`` under the key
+``"claude-code"``.
+
+Not yet included: subagent transcripts, which recent Claude Code versions write
+to separate files under ``<session-uuid>/subagents/``. Their spend is not part of
+the parent session's totals.
+"""
+
+import os
+import re
+from pathlib import Path
+
+from spooling.parser import ParsedSession, parse_session_file
+from spooling.providers.base import Provider
+
+_UUID_RE = re.compile(r"^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$")
+
+
+def _claude_config_dirs() -> list[Path]:
+    env = os.environ.get("CLAUDE_CONFIG_DIR", "").strip()
+    if env:
+        return [Path(env).expanduser()]
+    return [Path.home() / ".claude", Path.home() / ".config" / "claude"]
+
+
+class ClaudeCodeProvider(Provider):
+    type_id = "claude-code"
+    name = "Claude Code"
+
+    def default_data_path(self) -> Path:
+        candidates = [d / "projects" for d in _claude_config_dirs()]
+        for c in candidates:
+            if c.exists():
+                return c
+        return candidates[0]
+
+    def discover_session_files(self, data_path: Path | None = None) -> list[Path]:
+        base = data_path or self.resolved_data_path()
+        if not base.exists():
+            return []
+        files: list[Path] = []
+        for project_dir in base.iterdir():
+            if not project_dir.is_dir():
+                continue
+            # Top level only and UUID-named: skips subagent transcripts and
+            # non-session JSONL such as plugin skill-injection logs.
+            for f in project_dir.glob("*.jsonl"):
+                if _UUID_RE.match(f.stem):
+                    files.append(f)
+        return sorted(files, key=lambda f: f.stat().st_mtime, reverse=True)
+
+    def parse_session_file(self, file_path: Path) -> list[ParsedSession]:
+        session = parse_session_file(file_path, provider_id=self.type_id, project_from_cwd=True)
+        return [session] if session else []

{spooling-0.2.8 → spooling-0.2.9}/spooling/server.py

@@ -108,6 +108,13 @@ async def api_search(
 # --- Providers / Connections ---
 
 PROVIDER_TEMPLATES = {
+    "claude-code": {
+        "name": "Claude Code",
+        "icon": "claude",
+        "default_path": "~/.claude/projects",
+        "description": "Anthropic's Claude Code CLI. One JSONL transcript per session, with per-message token and prompt-cache usage.",
+        "status_hint": "Auto-detected from ~/.claude/projects/",
+    },
     "jsonl-session": {
         "name": "Session Files",
         "icon": "file",

{spooling-0.2.8 → spooling-0.2.9/spooling.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: spooling
-Version: 0.2.8
+Version: 0.2.9
 Summary: Local session tracker and semantic search for AI coding assistants
 Author: Parsed Analytics, Inc.
 License: MIT
@@ -38,7 +38,7 @@ Dynamic: license-file
 
 Local session tracker and semantic search for AI coding assistants.
 
-Track your AI coding sessions across **OpenAI Codex CLI**, **GitHub Copilot**, **Cursor**, **Windsurf**, **Kiro**, **Google Antigravity**, and **opencode**, all in one place. Get usage stats, cost estimates, per-provider breakdowns, semantic search, and a built-in AI chat agent to explore your history.
+Track your AI coding sessions across **Claude Code**, **OpenAI Codex CLI**, **GitHub Copilot**, **Cursor**, **Windsurf**, **Kiro**, **Google Antigravity**, and **opencode**, all in one place. Get usage stats, cost estimates, per-provider breakdowns, semantic search, and a built-in AI chat agent to explore your history.
 
 **Website:** [spooling.ai](https://spooling.ai)
 
@@ -142,6 +142,7 @@ spooling sync                      # Full sync with embeddings
 spooling sync --no-embed           # Skip embeddings (faster initial sync)
 spooling sync -p cursor            # Sync only Cursor sessions
 spooling sync -p codex             # Sync only Codex CLI sessions
+spooling sync -p claude-code       # Sync only Claude Code sessions
 ```
 
 ### `spooling stats`
@@ -639,6 +640,7 @@ All optional — defaults work out of the box.
 
 | Provider | Data Location | Format |
 |----------|--------------|--------|
+| **Claude Code** | `~/.claude/projects/` (or `$CLAUDE_CONFIG_DIR/projects`) | `<project>/<session-uuid>.jsonl` transcripts with per-message token usage, prompt-cache reads/writes, tool calls and git context |
 | **JSONL Sessions** | `~/.sessions/projects/` | UUID-named JSONL files with conversation history, tool calls, git context |
 | **OpenAI Codex CLI** | `~/.codex/sessions/` | `rollout-*.jsonl` files organized by date |
 | **GitHub Copilot** | `~/Library/Application Support/Code/User/workspaceStorage/` | Chat session JSON from VS Code |
@@ -650,5 +652,7 @@ All optional — defaults work out of the box.
 
 Run `spooling init` to see which providers are detected on your system.
 
+**Claude Code notes.** Sessions are detected automatically; there is nothing to connect. Cost is computed from each message's real token usage (including prompt-cache reads and writes) at the model's list price, so it reflects API-equivalent spend, not what a subscription plan charges. The project name is the session's working directory. To read transcripts from somewhere else, set `CLAUDE_CONFIG_DIR` or add `{"claude-code": "/path/to/projects"}` to `~/.config/spooling/paths.json`. Subagent transcripts (stored in separate files under `<session>/subagents/` by recent Claude Code versions) are not included yet.
+
 Nothing leaves your machine unless you opt in: Spooling Cloud (`spooling push`), a tunnel (`spooling tunnel`), or the Anthropic chat provider. `spooling pricing refresh` downloads public model prices but sends none of your data.
 

{spooling-0.2.8 → spooling-0.2.9}/spooling.egg-info/SOURCES.txt

@@ -51,6 +51,7 @@ spooling/connectors/rest/shopify_connectors.py
 spooling/providers/__init__.py
 spooling/providers/antigravity.py
 spooling/providers/base.py
+spooling/providers/claude_code.py
 spooling/providers/codex.py
 spooling/providers/copilot.py
 spooling/providers/cortex_code.py
@@ -62,6 +63,7 @@ spooling/providers/kiro.py
 spooling/providers/opencode.py
 spooling/providers/session_file.py
 spooling/providers/windsurf.py
+tests/test_claude_code_provider.py
 tests/test_mcp_auth_middleware.py
 tests/test_session_ids.py
 tests/test_storage_optimizations.py

spooling-0.2.9/tests/test_claude_code_provider.py

@@ -0,0 +1,229 @@
+"""Tests for the Claude Code provider (~/.claude/projects transcripts)."""
+
+import json
+import os
+import time
+from datetime import datetime, timezone
+
+import pytest
+
+from spooling import db as db_mod
+from spooling import ingest
+from spooling.parser import parse_session_file
+from spooling.providers import PROVIDER_REGISTRY, get_provider
+from spooling.providers import base as base_mod
+from spooling.providers.claude_code import ClaudeCodeProvider
+
+SID1 = "11111111-aaaa-4bbb-8ccc-000000000001"
+SID2 = "22222222-aaaa-4bbb-8ccc-000000000002"
+CWD = "/Users/dev/code/my-app"
+
+
+def _rec(type_, uuid, parent, ts, **kw):
+    r = {
+        "type": type_, "uuid": uuid, "parentUuid": parent, "isSidechain": False,
+        "timestamp": ts, "cwd": CWD, "gitBranch": "main", "version": "2.1.0",
+        "sessionId": SID1,
+    }
+    r.update(kw)
+    return r
+
+
+def _usage(i, o, cr=0, cw=0):
+    return {"input_tokens": i, "output_tokens": o,
+            "cache_read_input_tokens": cr, "cache_creation_input_tokens": cw}
+
+
+def transcript(sid: str = SID1) -> list[dict]:
+    """A realistic session: placeholder msg first, then prompt, tool use, answer."""
+    recs = [
+        {"type": "permission-mode", "permissionMode": "default", "sessionId": SID1},
+        {"type": "file-history-snapshot", "messageId": "x", "snapshot": {}},
+        # Locally generated placeholder: model "<synthetic>" must not become the session model.
+        _rec("assistant", "a0", None, "2026-11-01T10:00:00Z",
+             message={"role": "assistant", "model": "<synthetic>",
+                      "content": [{"type": "text", "text": "No response requested."}],
+                      "usage": _usage(0, 0)}),
+        _rec("user", "u1", "a0", "2026-11-01T10:00:05Z",
+             message={"role": "user", "content": "Fix the login bug in auth.ts"}),
+        _rec("assistant", "a1", "u1", "2026-11-01T10:00:10Z",
+             message={"role": "assistant", "model": "claude-sonnet-4-5",
+                      "content": [
+                          {"type": "text", "text": "Let me look at the file."},
+                          {"type": "tool_use", "id": "toolu_1", "name": "Read",
+                           "input": {"file_path": "/Users/dev/code/my-app/auth.ts"}}],
+                      "usage": _usage(100, 50, cr=2000, cw=500)}),
+        _rec("user", "u2", "a1", "2026-11-01T10:00:12Z",
+             message={"role": "user", "content": [
+                 {"type": "tool_result", "tool_use_id": "toolu_1", "content": "export function login() {}"}]}),
+        _rec("assistant", "a2", "u2", "2026-11-01T10:00:20Z",
+             message={"role": "assistant", "model": "claude-sonnet-4-5",
+                      "content": [{"type": "text", "text": "Fixed."}],
+                      "usage": _usage(10, 20, cr=2600, cw=0)}),
+        {"type": "ai-title", "title": "Fix login", "sessionId": SID1},
+    ]
+    # Message uuids are globally unique in real transcripts; keep the fixture honest.
+    for r in recs:
+        for k in ("uuid", "parentUuid"):
+            if r.get(k):
+                r[k] = f"{sid[:8]}-{r[k]}"
+        if "sessionId" in r:
+            r["sessionId"] = sid
+    return recs
+
+
+def write_session(root, project_dir, session_id, records, mtime=None):
+    d = root / "projects" / project_dir
+    d.mkdir(parents=True, exist_ok=True)
+    f = d / f"{session_id}.jsonl"
+    f.write_text("\n".join(json.dumps(r) for r in records) + "\n")
+    if mtime:
+        os.utime(f, (mtime, mtime))
+    return f
+
+
+@pytest.fixture
+def claude_home(tmp_path, monkeypatch):
+    home = tmp_path / "claude"
+    monkeypatch.setenv("CLAUDE_CONFIG_DIR", str(home))
+    # No user path overrides leaking in from the dev machine.
+    monkeypatch.setattr(base_mod, "PATH_OVERRIDES_FILE", tmp_path / "paths.json")
+    return home
+
+
+def test_registered():
+    assert PROVIDER_REGISTRY["claude-code"] is ClaudeCodeProvider
+    assert get_provider("claude-code").name == "Claude Code"
+
+
+def test_default_path_follows_claude_config_dir(claude_home):
+    assert ClaudeCodeProvider().default_data_path() == claude_home / "projects"
+
+
+def test_default_path_is_dot_claude_without_env(tmp_path, monkeypatch):
+    monkeypatch.delenv("CLAUDE_CONFIG_DIR", raising=False)
+    monkeypatch.setattr("pathlib.Path.home", lambda: tmp_path)
+    assert ClaudeCodeProvider().default_data_path() == tmp_path / ".claude" / "projects"
+
+
+def test_falls_back_to_xdg_location_when_dot_claude_missing(tmp_path, monkeypatch):
+    monkeypatch.delenv("CLAUDE_CONFIG_DIR", raising=False)
+    monkeypatch.setattr("pathlib.Path.home", lambda: tmp_path)
+    (tmp_path / ".config" / "claude" / "projects").mkdir(parents=True)
+    assert ClaudeCodeProvider().default_data_path() == tmp_path / ".config" / "claude" / "projects"
+
+
+def test_paths_json_override_wins(claude_home, tmp_path, monkeypatch):
+    custom = tmp_path / "elsewhere"
+    (tmp_path / "paths.json").write_text(json.dumps({"claude-code": str(custom)}))
+    assert ClaudeCodeProvider().resolved_data_path() == custom
+
+
+def test_not_available_when_directory_missing(claude_home):
+    assert ClaudeCodeProvider().is_available() is False
+    assert ClaudeCodeProvider().discover_session_files() == []
+
+
+def test_discovery_only_top_level_uuid_transcripts(claude_home):
+    old = write_session(claude_home, "-Users-dev-code-my-app", SID1, transcript(), mtime=time.time() - 1000)
+    new = write_session(claude_home, "-Users-dev-code-other", SID2, transcript(SID2))
+    proj = claude_home / "projects" / "-Users-dev-code-my-app"
+    # Things that must NOT be treated as sessions:
+    (proj / SID1 / "subagents").mkdir(parents=True)
+    (proj / SID1 / "subagents" / "agent-abc123.jsonl").write_text("{}\n")
+    (proj / "skill-injections.jsonl").write_text("{}\n")          # non-uuid name
+    (claude_home / "projects" / "stray.jsonl").write_text("{}\n")  # file directly in projects/
+
+    files = ClaudeCodeProvider().discover_session_files()
+    assert files == [new, old]  # newest first, nothing else
+
+
+def test_parse_stamps_provider_on_session_and_trace(claude_home):
+    f = write_session(claude_home, "-Users-dev-code-my-app", SID1, transcript())
+    [s] = ClaudeCodeProvider().parse_session_file(f)
+    assert s.provider_id == "claude-code"
+    assert s.trace is not None and s.trace.provider_id == "claude-code"
+    assert s.session_id == SID1
+
+
+def test_project_is_working_directory_name(claude_home):
+    f = write_session(claude_home, "-Users-dev-code-my-app", SID1, transcript())
+    [s] = ClaudeCodeProvider().parse_session_file(f)
+    assert s.project == "my-app"          # not "-Users-dev-code-my-app"
+    assert s.cwd == CWD and s.git_branch == "main" and s.agent_version == "2.1.0"
+
+
+def test_synthetic_placeholder_is_not_the_session_model(claude_home):
+    f = write_session(claude_home, "-Users-dev-code-my-app", SID1, transcript())
+    [s] = ClaudeCodeProvider().parse_session_file(f)
+    assert s.model == "claude-sonnet-4-5"
+
+
+def test_messages_tools_and_title(claude_home):
+    f = write_session(claude_home, "-Users-dev-code-my-app", SID1, transcript())
+    [s] = ClaudeCodeProvider().parse_session_file(f)
+    assert [m.role for m in s.messages] == ["assistant", "user", "assistant", "assistant"]
+    assert s.tool_call_count == 1
+    read = next(m for m in s.messages if m.tool_details).tool_details[0]
+    assert read.name == "Read" and read.input_summary.endswith("auth.ts")
+    assert read.result_preview == "export function login() {}"  # paired from the tool_result
+    assert s.title == "Fix the login bug in auth.ts"
+    assert s.started_at == datetime(2026, 11, 1, 10, 0, 0, tzinfo=timezone.utc)
+
+
+def test_token_usage_including_cache_flows_into_trace(claude_home):
+    f = write_session(claude_home, "-Users-dev-code-my-app", SID1, transcript())
+    [s] = ClaudeCodeProvider().parse_session_file(f)
+    llm = [sp for sp in s.trace.spans if getattr(sp.kind, "value", sp.kind) == "llm_call"]
+    assert len(llm) == 3
+    assert sum(x.input_tokens for x in llm) == 110
+    assert sum(x.output_tokens for x in llm) == 70
+    assert sum(x.cache_read_tokens for x in llm) == 4600
+    assert sum(x.cache_write_tokens for x in llm) == 500
+    # Cost uses all four components at claude-sonnet-4-5 rates (3 / 15 / 3.75 / 0.30 per Mtok).
+    real = [x for x in llm if x.model == "claude-sonnet-4-5"]
+    expected = (110 * 3 + 70 * 15 + 500 * 3.75 + 4600 * 0.30) / 1e6
+    assert sum(x.cost_usd for x in real) == pytest.approx(expected, rel=0.02)
+
+
+def test_empty_or_garbage_files_yield_no_session(claude_home):
+    f = write_session(claude_home, "-p", SID2, [{"type": "permission-mode"}])
+    assert ClaudeCodeProvider().parse_session_file(f) == []
+    g = claude_home / "projects" / "-p" / f"{SID1}.jsonl"
+    g.write_text("not json\n\n{broken\n")
+    assert ClaudeCodeProvider().parse_session_file(g) == []
+
+
+def test_generic_session_file_parser_is_unchanged(claude_home):
+    """The legacy jsonl-session path keeps its id and folder-name project."""
+    f = write_session(claude_home, "-Users-dev-code-my-app", SID1, transcript())
+    s = parse_session_file(f)
+    assert s.provider_id == "jsonl-session" and s.trace.provider_id == "jsonl-session"
+    assert s.project == "-Users-dev-code-my-app"
+
+
+def test_sync_ingests_and_is_idempotent(claude_home, tmp_path, monkeypatch):
+    monkeypatch.setattr(db_mod, "DB_PATH", tmp_path / "spooling.db")
+    write_session(claude_home, "-Users-dev-code-my-app", SID1, transcript())
+    write_session(claude_home, "-Users-dev-code-other", SID2, transcript(SID2))
+
+    ingest.sync(embed=False, provider_filter="claude-code")
+
+    conn = db_mod.get_connection()
+    rows = conn.execute("SELECT id, provider_id, project, model FROM sessions ORDER BY id").fetchall()
+    assert [(r["id"], r["provider_id"], r["project"], r["model"]) for r in rows] == [
+        (SID1, "claude-code", "my-app", "claude-sonnet-4-5"),
+        (SID2, "claude-code", "my-app", "claude-sonnet-4-5"),
+    ]
+    assert conn.execute("SELECT COUNT(*) AS n FROM messages").fetchone()["n"] == 8
+    assert [r["provider_id"] for r in conn.execute("SELECT DISTINCT provider_id FROM traces").fetchall()] == ["claude-code"]
+    spans = conn.execute("SELECT COUNT(*) AS n FROM spans WHERE kind = 'llm_call'").fetchone()["n"]
+    assert spans == 6
+    conn.close()
+
+    # Second run: nothing changed on disk, nothing re-ingested.
+    ingest.sync(embed=False, provider_filter="claude-code")
+    conn = db_mod.get_connection()
+    assert conn.execute("SELECT COUNT(*) AS n FROM sessions").fetchone()["n"] == 2
+    assert conn.execute("SELECT COUNT(*) AS n FROM spans WHERE kind = 'llm_call'").fetchone()["n"] == 6
+    conn.close()