nabla-cli 0.7.0__tar.gz → 0.7.2__tar.gz

{nabla_cli-0.7.0 → nabla_cli-0.7.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nabla-cli
-Version: 0.7.0
+Version: 0.7.2
 Summary: Call-site harvester: scan OpenAI call sites, record real traffic, emit site profiles
 License: MIT
 Requires-Python: >=3.11
@@ -170,14 +170,15 @@ a problem generator:
 
 ```json
 {
-  "goal": "You are generating new problems to test and eventually fine-tune a small student model on the following task: ...",
+  "goal": "<build's inferred goal description and constraints, joined into one string>",
   "system_prompt": "<the site's own system prompt>",
   "examples": [ {"problem": "...", "solution": "..."}, ... ]
 }
 ```
 
-The goal text is composed deterministically from the same inferred goal
-and constraints that `build` uses; the examples are the captured pairs.
+Same pipeline as `build` (captured pairs in, one goal-inference call) —
+only the output shape differs; nothing is added that `build`'s own
+output doesn't contain.
 
 ## Providers
 

{nabla_cli-0.7.0 → nabla_cli-0.7.2}/README.md

@@ -156,14 +156,15 @@ a problem generator:
 
 ```json
 {
-  "goal": "You are generating new problems to test and eventually fine-tune a small student model on the following task: ...",
+  "goal": "<build's inferred goal description and constraints, joined into one string>",
   "system_prompt": "<the site's own system prompt>",
   "examples": [ {"problem": "...", "solution": "..."}, ... ]
 }
 ```
 
-The goal text is composed deterministically from the same inferred goal
-and constraints that `build` uses; the examples are the captured pairs.
+Same pipeline as `build` (captured pairs in, one goal-inference call) —
+only the output shape differs; nothing is added that `build`'s own
+output doesn't contain.
 
 ## Providers
 

{nabla_cli-0.7.0 → nabla_cli-0.7.2}/nabla/__init__.py

@@ -1,3 +1,3 @@
 """nabla — call-site harvester CLI (pre-Phase-0)."""
 
-__version__ = "0.7.0"
+__version__ = "0.7.2"

{nabla_cli-0.7.0 → nabla_cli-0.7.2}/nabla/cli.py

@@ -137,9 +137,10 @@ def _build_parser() -> argparse.ArgumentParser:
         help="assemble a seed spec (goal + system_prompt + problem/solution examples)",
         formatter_class=argparse.RawDescriptionHelpFormatter,
         epilog=(
-            "an alternative to `build`: same inputs, but emits <site>.jeremy.json —\n"
-            "a generator-framed goal, the site's system prompt, and the captured\n"
-            "pairs as {problem, solution} reference examples.\n\n"
+            "an alternative to `build`: same pipeline and inputs, but emits\n"
+            "<site>.jeremy.json — build's goal rearranged into one string, the\n"
+            "site's system prompt as its own field, and the captured pairs as\n"
+            "{problem, solution} reference examples.\n\n"
             "examples:\n"
             "  nabla jeremy                       auto: pick site, find <site>.captured.jsonl\n"
             "  nabla jeremy path/to/repo --site extract_invoice --captured captured.jsonl\n"
@@ -568,9 +569,11 @@ def _cmd_record(args) -> int:
                    "data and rerun for a higher-fidelity harvest", style="yellow")
     rail(_out)
     if args.out is None and args.site is None:
-        rail_end(_out, f"next: nabla build{'' if args.repo == '.' else ' ' + args.repo}")
+        rail_end(_out, f"next: nabla build{'' if args.repo == '.' else ' ' + args.repo}"
+                       " (site profile) · or: nabla jeremy (seed spec)")
     else:
-        rail_end(_out, f"next: nabla build {args.repo} --site {site.symbol} --captured {out}")
+        rail_end(_out, f"next: nabla build {args.repo} --site {site.symbol} --captured {out}"
+                       " · or: nabla jeremy (same flags)")
     return 0
 
 
@@ -762,7 +765,8 @@ def _capture_many(args, supported) -> int:
            for r in failed):
         _warn("the configured provider rejected your API key — rerun `nabla init`")
     rail(_out)
-    rail_end(_out, f"next: nabla build{'' if args.repo == '.' else ' ' + args.repo}")
+    rail_end(_out, f"next: nabla build{'' if args.repo == '.' else ' ' + args.repo}"
+                   " (site profile) · or: nabla jeremy (seed spec)")
     return 1 if len(failed) == len(results) else 0
 
 
@@ -844,17 +848,11 @@ def _cmd_profile(args, mode: str = "profile") -> int:
 
 
 def _compose_seed_goal(goal) -> str:
-    """Generator-framed goal text for the seed spec, composed in pure Python
-    from the induced GoalBlock — no extra LLM call, no invented facts."""
-    parts = [
-        "You are generating new problems to test and eventually fine-tune a "
-        "small student model on the following task: " + goal.description.rstrip(".") + "."
-    ]
-    if goal.constraints:
-        parts.append("Every generated problem's solution must satisfy: "
-                     + " ".join(c.rstrip(".") + "." for c in goal.constraints))
-    parts.append("Match the style, difficulty range, and output format of the "
-                 "reference examples below.")
+    """The seed spec's goal string: build's induced goal, rearranged into
+    prose — description followed by the constraints. Nothing is added that
+    build's own output doesn't contain."""
+    parts = [goal.description.rstrip(".") + "."]
+    parts += [c.rstrip(".") + "." for c in goal.constraints]
     return " ".join(parts)
 
 

{nabla_cli-0.7.0 → nabla_cli-0.7.2}/nabla_cli.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nabla-cli
-Version: 0.7.0
+Version: 0.7.2
 Summary: Call-site harvester: scan OpenAI call sites, record real traffic, emit site profiles
 License: MIT
 Requires-Python: >=3.11
@@ -170,14 +170,15 @@ a problem generator:
 
 ```json
 {
-  "goal": "You are generating new problems to test and eventually fine-tune a small student model on the following task: ...",
+  "goal": "<build's inferred goal description and constraints, joined into one string>",
   "system_prompt": "<the site's own system prompt>",
   "examples": [ {"problem": "...", "solution": "..."}, ... ]
 }
 ```
 
-The goal text is composed deterministically from the same inferred goal
-and constraints that `build` uses; the examples are the captured pairs.
+Same pipeline as `build` (captured pairs in, one goal-inference call) —
+only the output shape differs; nothing is added that `build`'s own
+output doesn't contain.
 
 ## Providers
 

{nabla_cli-0.7.0 → nabla_cli-0.7.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "nabla-cli"
-version = "0.7.0"
+version = "0.7.2"
 description = "Call-site harvester: scan OpenAI call sites, record real traffic, emit site profiles"
 readme = "README.md"
 license = { text = "MIT" }

{nabla_cli-0.7.0 → nabla_cli-0.7.2}/tests/test_cli_ux.py

@@ -517,72 +517,71 @@ def test_scan_skips_build_and_dist_dirs(tmp_path, monkeypatch, capsys):
     out = capsys.readouterr().out
     assert "1 call site found" in out
     assert "copy_of_site" not in out
-
-
-# ---------------------------------------------------------------------------
-# nabla jeremy — seed-spec sibling of build
-# ---------------------------------------------------------------------------
-
-def _fake_goal(*a, **k):
-    from nabla.contract import GoalBlock
-
-    return GoalBlock(
-        description="Solve a grade-school word problem step by step.",
-        constraints=["final answer is a bare number", "no invented constants"],
-        difficulty_dimensions=[{"name": "d", "levels": ["a", "b"], "evidence": "e"}],
-    )
-
-
-def test_jeremy_single_site_writes_seed_spec(tmp_path, monkeypatch, capsys):
-    repo = _make_single_site_repo(tmp_path)
-    monkeypatch.chdir(tmp_path)
-    pair = {"input_slots": {"ocr_text": "r"}, "prompt": "A farmer has 6 crates...",
-            "completion": "The answer is 57.",
-            "tokens": {"in": 5, "out": 5}, "latency_ms": 10, "_source_kind": "samples"}
-    (tmp_path / "extract_receipt.captured.jsonl").write_text(
-        json.dumps(pair) + "\n", encoding="utf-8")
-    monkeypatch.setattr("nabla.induce.induce", _fake_goal)
-
-    rc = main(["jeremy", str(repo)])
-
-    assert rc == 0
-    seed = json.loads((tmp_path / "extract_receipt.jeremy.json").read_text(encoding="utf-8"))
-    assert set(seed) == {"goal", "system_prompt", "examples"}
-    assert seed["goal"].startswith("You are generating new problems")
-    assert "grade-school word problem" in seed["goal"]
-    assert "final answer is a bare number" in seed["goal"]
-    assert "Extract the receipt fields" in seed["system_prompt"]
-    assert seed["examples"] == [
-        {"problem": "A farmer has 6 crates...", "solution": "The answer is 57."}
-    ]
-    assert "extract_receipt.jeremy.json" in capsys.readouterr().out
-
-
-def test_jeremy_multi_site(tmp_path, monkeypatch, capsys):
-    repo = _make_two_site_repo(tmp_path)
-    monkeypatch.chdir(tmp_path)
-    pair = {"input_slots": {}, "prompt": "p", "completion": "s",
-            "tokens": {"in": 1, "out": 1}, "latency_ms": 1, "_source_kind": "samples"}
-    for symbol in ("extract_receipt", "extract_menu"):
-        (tmp_path / f"{symbol}.captured.jsonl").write_text(
-            json.dumps(pair) + "\n", encoding="utf-8")
-    monkeypatch.setattr("nabla.induce.induce", _fake_goal)
-
-    rc = main(["jeremy", str(repo)])
-
-    assert rc == 0
-    assert (tmp_path / "extract_receipt.jeremy.json").exists()
-    assert (tmp_path / "extract_menu.jeremy.json").exists()
-    assert "built 2 seed spec(s)" in capsys.readouterr().out
-
-
-def test_jeremy_without_captures_points_at_capture(tmp_path, monkeypatch, capsys):
-    repo = _make_single_site_repo(tmp_path)
-    monkeypatch.chdir(tmp_path)
-
-    rc = main(["jeremy", str(repo)])
-
-    assert rc == 2
-    err = capsys.readouterr().err
-    assert "nabla jeremy" in err
-    assert "nabla capture" in err
+
+
+# ---------------------------------------------------------------------------
+# nabla jeremy — seed-spec sibling of build
+# ---------------------------------------------------------------------------
+
+def _fake_goal(*a, **k):
+    from nabla.contract import GoalBlock
+
+    return GoalBlock(
+        description="Solve a grade-school word problem step by step.",
+        constraints=["final answer is a bare number", "no invented constants"],
+        difficulty_dimensions=[{"name": "d", "levels": ["a", "b"], "evidence": "e"}],
+    )
+
+
+def test_jeremy_single_site_writes_seed_spec(tmp_path, monkeypatch, capsys):
+    repo = _make_single_site_repo(tmp_path)
+    monkeypatch.chdir(tmp_path)
+    pair = {"input_slots": {"ocr_text": "r"}, "prompt": "A farmer has 6 crates...",
+            "completion": "The answer is 57.",
+            "tokens": {"in": 5, "out": 5}, "latency_ms": 10, "_source_kind": "samples"}
+    (tmp_path / "extract_receipt.captured.jsonl").write_text(
+        json.dumps(pair) + "\n", encoding="utf-8")
+    monkeypatch.setattr("nabla.induce.induce", _fake_goal)
+
+    rc = main(["jeremy", str(repo)])
+
+    assert rc == 0
+    seed = json.loads((tmp_path / "extract_receipt.jeremy.json").read_text(encoding="utf-8"))
+    assert set(seed) == {"goal", "system_prompt", "examples"}
+    assert seed["goal"].startswith("Solve a grade-school word problem")
+    assert "final answer is a bare number" in seed["goal"]
+    assert "Extract the receipt fields" in seed["system_prompt"]
+    assert seed["examples"] == [
+        {"problem": "A farmer has 6 crates...", "solution": "The answer is 57."}
+    ]
+    assert "extract_receipt.jeremy.json" in capsys.readouterr().out
+
+
+def test_jeremy_multi_site(tmp_path, monkeypatch, capsys):
+    repo = _make_two_site_repo(tmp_path)
+    monkeypatch.chdir(tmp_path)
+    pair = {"input_slots": {}, "prompt": "p", "completion": "s",
+            "tokens": {"in": 1, "out": 1}, "latency_ms": 1, "_source_kind": "samples"}
+    for symbol in ("extract_receipt", "extract_menu"):
+        (tmp_path / f"{symbol}.captured.jsonl").write_text(
+            json.dumps(pair) + "\n", encoding="utf-8")
+    monkeypatch.setattr("nabla.induce.induce", _fake_goal)
+
+    rc = main(["jeremy", str(repo)])
+
+    assert rc == 0
+    assert (tmp_path / "extract_receipt.jeremy.json").exists()
+    assert (tmp_path / "extract_menu.jeremy.json").exists()
+    assert "built 2 seed spec(s)" in capsys.readouterr().out
+
+
+def test_jeremy_without_captures_points_at_capture(tmp_path, monkeypatch, capsys):
+    repo = _make_single_site_repo(tmp_path)
+    monkeypatch.chdir(tmp_path)
+
+    rc = main(["jeremy", str(repo)])
+
+    assert rc == 2
+    err = capsys.readouterr().err
+    assert "nabla jeremy" in err
+    assert "nabla capture" in err