diff --git a/README.md b/README.md
index f4d990d..2290163 100644
--- a/README.md
+++ b/README.md
@@ -199,6 +199,10 @@ uv run probelock probe  --tools tools.json --traces traces.json \
     --endpoint http://localhost:11434/v1 --model llama3.1:8b -o candidate.lock
 ```
 
+`--tools` is optional here: traced probes replay their own embedded tool definitions,
+so a trace-only run (`probe --traces traces.json ...`) needs no schema file. The same
+holds for `--mined` below. Provide `--tools` when you also want the synthetic battery.
+
 A traces file is a small, stable JSON schema probelock defines itself — **not** raw
 OpenTelemetry — because OTel's own span attribute layout is not stable across libraries or
 versions (litellm has already changed where it puts request/response attributes once, and
diff --git a/VALIDATION.md b/VALIDATION.md
index 32ca5ac..b4ba4e3 100644
--- a/VALIDATION.md
+++ b/VALIDATION.md
@@ -264,8 +264,9 @@ force the code path this test targets.
 - `llama-server -fa` requires an explicit value (`--flash-attn auto`) on
   recent builds; the bare flag misparses a following `-hf` argument as its
   value.
-- `probelock probe --traces` still requires `--tools`; the traced probes carry
-  their own embedded tool definitions regardless of what's in the schema file.
+- `probelock probe --traces` required `--tools` at the time of this validation,
+  even though traced probes carry their own embedded tool definitions. Fixed in
+  v0.3.1: `--tools` is now optional whenever `--traces` or `--mined` supply probes.
 - Running two `llama-server` instances concurrently on one GPU crashed one of
   them (Metal contention) mid-probe during testing. Don't run concurrent GPU
   inference workloads on a single-GPU host.
diff --git a/probelock/__init__.py b/probelock/__init__.py
index a34bb6c..bfc6dab 100644
--- a/probelock/__init__.py
+++ b/probelock/__init__.py
@@ -5,4 +5,4 @@
 model version, quantization, or runtime.
 """
 
-__version__ = "0.3.0"
+__version__ = "0.3.1"
diff --git a/probelock/cli.py b/probelock/cli.py
index de1f38b..095527e 100644
--- a/probelock/cli.py
+++ b/probelock/cli.py
@@ -160,14 +160,27 @@ def _mined_battery(path: Path):
     return accepted, [to_probe(p) for p in accepted]
 
 
+_TOOLS_OPTION = typer.Option(
+    None, "--tools", "-t", help="OpenAI-style tools JSON file (optional when --traces "
+    "or --mined supply probes — those carry their own embedded tool definitions)."
+)
+
+
+def _require_probe_source(tools, traces, mined) -> None:
+    if tools is None and traces is None and mined is None:
+        _err("Provide at least one probe source: --tools, --traces, or --mined.")
+        raise typer.Exit(2)
+
+
 @app.command()
 def derive(
-    tools: Path = typer.Option(..., "--tools", "-t", help="OpenAI-style tools JSON file."),
+    tools: Optional[Path] = _TOOLS_OPTION,
     traces: Optional[Path] = _TRACES_OPTION,
     mined: Optional[Path] = _MINED_OPTION,
 ) -> None:
     """Show the probe battery that would be generated from a toolset (transparency)."""
-    probes = derive_probes(_load_tools_or_exit(tools))
+    _require_probe_source(tools, traces, mined)
+    probes = derive_probes(_load_tools_or_exit(tools)) if tools is not None else []
     if traces is not None:
         probes = probes + derive_traced_probes(_load_traces_or_exit(traces))
     if mined is not None:
@@ -183,7 +196,7 @@ def derive(
 
 @app.command()
 def probe(
-    tools: Path = typer.Option(..., "--tools", "-t", help="OpenAI-style tools JSON file."),
+    tools: Optional[Path] = _TOOLS_OPTION,
     simulate: Optional[Path] = typer.Option(
         None, "--simulate", "-s", help="Run against a deterministic profile (no model)."
     ),
@@ -215,9 +228,18 @@ def probe(
     ),
 ) -> None:
     """Run the probe battery and produce a capability lockfile."""
-    tool_list = _load_tools_or_exit(tools)
-    probes = derive_probes(tool_list)
-    fingerprint = tools_fingerprint(tool_list)
+    _require_probe_source(tools, traces, mined)
+    if tools is not None:
+        tool_list = _load_tools_or_exit(tools)
+        probes = derive_probes(tool_list)
+        fingerprint = tools_fingerprint(tool_list)
+    else:
+        # Trace-only run: traced/mined probes replay their own embedded tool
+        # definitions, so there is no schema battery and no toolset to fingerprint.
+        # The empty fingerprint is stable, so two trace-only lockfiles diff cleanly,
+        # while a trace-only vs schema-derived pair still flags tools_changed.
+        probes = []
+        fingerprint = ""
 
     traces_fp = None
     if traces is not None:
@@ -264,6 +286,12 @@ def probe(
     elif fps:
         traces_fp = fps[0]
 
+    if not probes:
+        # Possible without --tools: a traces file that derives nothing, or a mined
+        # file with no accepted probes. A zero-probe lockfile can gate nothing.
+        _err("The probe battery is empty — nothing derived from the given sources.")
+        raise typer.Exit(2)
+
     try:
         if simulate is not None:
             client = SimulatedClient(_load_json_or_exit(simulate, "simulate profile"))
diff --git a/pyproject.toml b/pyproject.toml
index b583a7a..2156b81 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -1,6 +1,6 @@
 [project]
 name = "probelock"
-version = "0.3.0"
+version = "0.3.1"
 description = "A capability lockfile for local models. Catch silent regressions when you swap a model, quant, or runtime."
 readme = "README.md"
 requires-python = ">=3.10"
diff --git a/tests/test_cli.py b/tests/test_cli.py
index 41d626b..4a2711c 100644
--- a/tests/test_cli.py
+++ b/tests/test_cli.py
@@ -588,3 +588,55 @@ def test_ingest_accepts_multiple_logs_as_one_corpus(tmp_path):
     assert cats.count("tool_selection") == 2  # incl. the continuation across files
     assert cats.count("schema_validity") == 5
     assert cats.count("no_tool") == 1
+
+
+# --- --tools optional when traces/mined supply probes (validation rough edge) -------
+
+
+def test_probe_traces_only_needs_no_tools_file(tmp_path):
+    lock = tmp_path / "traced.lock"
+    result = runner.invoke(app, [
+        "probe", "--traces", str(_SAMPLE_TRACES), "--simulate", str(_SIM_Q8),
+        "-o", str(lock)])
+    assert result.exit_code == 0, result.output
+    data = json.loads(lock.read_text())
+    assert data["tools_fingerprint"] == ""  # no schema battery ran
+    assert data["traces_fingerprint"]
+    assert "tool_selection" in data["capabilities"]  # traced probes replayed
+    assert data["n_probes"] > 0
+
+
+def test_derive_traces_only_needs_no_tools_file():
+    result = runner.invoke(app, ["derive", "--traces", str(_SAMPLE_TRACES)])
+    assert result.exit_code == 0, result.output
+    assert "::traced::" in result.output
+
+
+def test_probe_mined_only_needs_no_tools_file(tmp_path):
+    mined = _ingest(tmp_path)
+    runner.invoke(app, ["traces", "review", str(mined), "--auto-accept", "schema_validity"])
+    lock = tmp_path / "mined-only.lock"
+    result = runner.invoke(app, [
+        "probe", "--mined", str(mined), "--simulate", str(_SIM_Q8),
+        "--allow-sensitive", "-o", str(lock)])
+    assert result.exit_code == 0, result.output
+    caps = json.loads(lock.read_text())["capabilities"]
+    assert list(caps) == ["traced_schema_validity"]
+
+
+def test_probe_without_any_source_exits_2():
+    result = runner.invoke(app, ["probe", "--simulate", str(_SIM_Q8)])
+    assert result.exit_code == 2, result.output
+    assert "--tools, --traces, or --mined" in result.output
+    result = runner.invoke(app, ["derive"])
+    assert result.exit_code == 2, result.output
+
+
+def test_probe_with_empty_battery_exits_2(tmp_path):
+    # a mined file with nothing accepted and no --tools: the battery is empty and a
+    # zero-probe lockfile can gate nothing
+    mined = _ingest(tmp_path)  # everything still pending
+    result = runner.invoke(app, [
+        "probe", "--mined", str(mined), "--simulate", str(_SIM_Q8)])
+    assert result.exit_code == 2, result.output
+    assert "empty" in result.output