nabla-cli 0.9.0__tar.gz → 0.9.2__tar.gz

{nabla_cli-0.9.0 → nabla_cli-0.9.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nabla-cli
-Version: 0.9.0
+Version: 0.9.2
 Summary: Call-site harvester: scan OpenAI call sites, record real traffic, emit site profiles
 License: MIT
 Requires-Python: >=3.11

{nabla_cli-0.9.0 → nabla_cli-0.9.2}/nabla/__init__.py

@@ -1,3 +1,3 @@
 """nabla — call-site harvester CLI (pre-Phase-0)."""
 
-__version__ = "0.9.0"
+__version__ = "0.9.2"

{nabla_cli-0.9.0 → nabla_cli-0.9.2}/nabla/cli.py

@@ -14,7 +14,7 @@ from rich.console import Console
 from . import __version__
 from .branding import print_banner
 from .config import PROVIDERS, apply_to_env, save_config
-from .contract import Example, Verifiability, is_supported
+from .contract import Example, Verifiability, compute_site_id, is_supported
 from .profile import build_profile, find_site
 from .recorder import RecordSource, ZeroTrafficError, record
 from .scanner import scan
@@ -947,6 +947,11 @@ def _build_one(args, site, final_rail: bool = False, mode: str = "profile") -> i
 
     if mode == "jeremy":
         seed = {
+            # Stable site identity (template+schema hash) — the thread that
+            # lets `pull` later match a fine-tuned deployment back to this
+            # exact call site. Same value as the profile's site_id.
+            "site_id": compute_site_id(template.template, schema),
+            "symbol": site.symbol,
             "goal": _compose_seed_goal(goal),
             "system_prompt": template.system_prompt or "",
             "examples": [
@@ -1083,6 +1088,12 @@ def _cmd_pull(args) -> int:
     for site in supported:
         replaces = extract_sampling(site, args.repo).get("model") or "unknown"
         deployments.append((site, mock_deployment(site.symbol, replaces)))
+    # Simulation richness: a lone site still demos a multi-row pull.
+    filler = None
+    if len(deployments) == 1 and args.site is None:
+        from .puller import filler_deployment
+        filler = filler_deployment()
+        deployments.append(filler)
 
     rail(_out)
     table = Table(box=rich_box.SIMPLE_HEAD, header_style="bold", pad_edge=False)
@@ -1098,10 +1109,14 @@ def _cmd_pull(args) -> int:
         )
     _out.print(table)
 
+    from .puller import filler_diff
+
     for site, dep in deployments:
         rail(_out)
         rail(_out, f"{site.symbol} · {site.file}", style="bold")
-        for style, line in build_diff(args.repo, site, dep):
+        is_filler = filler is not None and site is filler[0]
+        diff = filler_diff(dep) if is_filler else build_diff(args.repo, site, dep)
+        for style, line in diff:
             color = {"add": "green", "del": "red", "hunk": "cyan",
                      "head": "bold", "ctx": ""}[style]
             rail(_out, line, style=color)

{nabla_cli-0.9.0 → nabla_cli-0.9.2}/nabla/puller.py

@@ -1,119 +1,148 @@
-"""`nabla pull` — UX simulation of the loop-closing command (cli-plan §7).
-
-The deployments API doesn't exist yet, so this module renders the full
-intended experience — deployment table, per-site diffs, apply + verify,
-before/after report — as a faithful simulation that NEVER writes anything:
-no file edits, no .env changes, no network calls. Diffs are built from the
-target repo's real source lines so the preview is true to life; deployment
-rows are fabricated deterministically from the site symbol.
-
-When the real backend lands, the fabrication points (_mock_deployments,
-the apply/verify steps) get replaced by API calls and real edits; the
-rendering stays.
-"""
-
-from __future__ import annotations
-
-import os
-import time
-from pathlib import Path
-
-from .config import load_config
-
-DEFAULT_ENDPOINT = "http://127.0.0.1:8000/deployments"
-_BASE_URL = "https://api.freesolo.ai/v1"
-_KEY_ENV = "NABLA_MODEL_API_KEY"
-
-
-def resolve_endpoint(flag_value: str | None = None) -> str:
-    if flag_value:
-        return flag_value
-    env = os.environ.get("NABLA_PULL_ENDPOINT")
-    if env:
-        return env
-    stored = load_config().get("pull_endpoint")
-    if stored:
-        return stored
-    return DEFAULT_ENDPOINT
-
-
-def mock_deployment(symbol: str, replaces: str) -> dict:
-    """A deterministic, plausible deployment row for one site."""
-    slug = symbol.replace("_", "-")
-    seed = sum(ord(c) for c in symbol)
-    return {
-        "symbol": symbol,
-        "status": "ready",
-        "model": f"freesolo/qwen3.5-0.8b-ft-{slug}",
-        "base_url": _BASE_URL,
-        "api_key_env": _KEY_ENV,
-        "replaces": replaces or "unknown",
-        "pass_rate": round(0.88 + (seed % 10) / 100, 2),
-        "schema_valid_rate": 1.0,
-    }
-
-
-def build_diff(repo: str, site, deployment: dict) -> list[tuple[str, str]]:
-    """(style, line) pairs for a unified diff of the intended rewrite,
-    grounded in the real source. Read-only."""
-    path = Path(repo) / site.file
-    try:
-        lines = path.read_text(encoding="utf-8").splitlines()
-    except OSError:
-        return []
-
-    env_suffix = site.symbol.upper()
-    out: list[tuple[str, str]] = [
-        ("head", f"--- a/{site.file}"),
-        ("head", f"+++ b/{site.file}"),
-    ]
-
-    # Hunk 1: client construction, if a literal OpenAI(...) line exists.
-    client_idx = next((i for i, l in enumerate(lines)
-                       if "OpenAI(" in l and "import" not in l), None)
-    if client_idx is not None:
-        old = lines[client_idx]
-        indent = old[: len(old) - len(old.lstrip())]
-        assign = old.split("OpenAI(", 1)[0]
-        out += [
-            ("hunk", f"@@ -{client_idx + 1},1 +{client_idx + 1},2 @@"),
-            ("del", f"-{old}"),
-            ("add", f"+{assign}OpenAI("),
-            ("add", f"+{indent}    base_url=os.environ.get(\"NABLA_BASE_URL\", "
-                    f"\"{deployment['base_url']}\"),"),
-            ("add", f"+{indent}    api_key=os.environ[\"{deployment['api_key_env']}\"],"),
-            ("add", f"+{indent})"),
-        ]
-
-    # Hunk 2: the model kwarg inside the call span.
-    start, end = site.span
-    model_idx = next((i for i in range(start - 1, min(end, len(lines)))
-                      if "model=" in lines[i] or "model =" in lines[i]), None)
-    if model_idx is not None:
-        old = lines[model_idx]
-        indent = old[: len(old) - len(old.lstrip())]
-        trailing = "," if old.rstrip().endswith(",") else ""
-        context = lines[model_idx - 1] if model_idx else ""
-        out += [
-            ("hunk", f"@@ -{model_idx},2 +{model_idx},2 @@"),
-            ("ctx", f" {context}"),
-            ("del", f"-{old}"),
-            ("add", f"+{indent}model=os.environ.get(\"NABLA_MODEL_{env_suffix}\", "
-                    f"\"{deployment['model']}\"){trailing}"),
-        ]
-
-    # Hunk 3: credentials, appended to the repo's .env.
-    out += [
-        ("head", "--- a/.env"),
-        ("head", "+++ b/.env"),
-        ("add", f"+{deployment['api_key_env']}=fsk-................ (provisioned by backend)"),
-        ("add", f"+NABLA_BASE_URL={deployment['base_url']}"),
-        ("add", f"+NABLA_MODEL_{env_suffix}={deployment['model']}"),
-    ]
-    return out
-
-
-def pace(seconds: float, interactive: bool) -> None:
-    """Small dramatic pause, interactive terminals only."""
-    if interactive:
-        time.sleep(seconds)
+"""`nabla pull` — UX simulation of the loop-closing command (cli-plan §7).
+
+The deployments API doesn't exist yet, so this module renders the full
+intended experience — deployment table, per-site diffs, apply + verify,
+before/after report — as a faithful simulation that NEVER writes anything:
+no file edits, no .env changes, no network calls. Diffs are built from the
+target repo's real source lines so the preview is true to life; deployment
+rows are fabricated deterministically from the site symbol.
+
+When the real backend lands, the fabrication points (_mock_deployments,
+the apply/verify steps) get replaced by API calls and real edits; the
+rendering stays.
+"""
+
+from __future__ import annotations
+
+import os
+import time
+from pathlib import Path
+
+from .config import load_config
+
+DEFAULT_ENDPOINT = "http://127.0.0.1:8000/deployments"
+_BASE_URL = "https://api.freesolo.ai/v1"
+_KEY_ENV = "NABLA_MODEL_API_KEY"
+
+
+def resolve_endpoint(flag_value: str | None = None) -> str:
+    if flag_value:
+        return flag_value
+    env = os.environ.get("NABLA_PULL_ENDPOINT")
+    if env:
+        return env
+    stored = load_config().get("pull_endpoint")
+    if stored:
+        return stored
+    return DEFAULT_ENDPOINT
+
+
+def mock_deployment(symbol: str, replaces: str) -> dict:
+    """A deterministic, plausible deployment row for one site."""
+    slug = symbol.replace("_", "-")
+    seed = sum(ord(c) for c in symbol)
+    return {
+        "symbol": symbol,
+        "status": "ready",
+        "model": f"freesolo/qwen3.5-0.8b-ft-{slug}",
+        "base_url": _BASE_URL,
+        "api_key_env": _KEY_ENV,
+        "replaces": replaces or "unknown",
+        "pass_rate": round(0.88 + (seed % 10) / 100, 2),
+        "schema_valid_rate": 1.0,
+    }
+
+
+def build_diff(repo: str, site, deployment: dict) -> list[tuple[str, str]]:
+    """(style, line) pairs for a unified diff of the intended rewrite,
+    grounded in the real source. Read-only."""
+    path = Path(repo) / site.file
+    try:
+        lines = path.read_text(encoding="utf-8").splitlines()
+    except OSError:
+        return []
+
+    env_suffix = site.symbol.upper()
+    out: list[tuple[str, str]] = [
+        ("head", f"--- a/{site.file}"),
+        ("head", f"+++ b/{site.file}"),
+    ]
+
+    # Hunk 1: client construction, if a literal OpenAI(...) line exists.
+    client_idx = next((i for i, l in enumerate(lines)
+                       if "OpenAI(" in l and "import" not in l), None)
+    if client_idx is not None:
+        old = lines[client_idx]
+        indent = old[: len(old) - len(old.lstrip())]
+        assign = old.split("OpenAI(", 1)[0]
+        out += [
+            ("hunk", f"@@ -{client_idx + 1},1 +{client_idx + 1},2 @@"),
+            ("del", f"-{old}"),
+            ("add", f"+{assign}OpenAI("),
+            ("add", f"+{indent}    base_url=os.environ.get(\"NABLA_BASE_URL\", "
+                    f"\"{deployment['base_url']}\"),"),
+            ("add", f"+{indent}    api_key=os.environ[\"{deployment['api_key_env']}\"],"),
+            ("add", f"+{indent})"),
+        ]
+
+    # Hunk 2: the model kwarg inside the call span.
+    start, end = site.span
+    model_idx = next((i for i in range(start - 1, min(end, len(lines)))
+                      if "model=" in lines[i] or "model =" in lines[i]), None)
+    if model_idx is not None:
+        old = lines[model_idx]
+        indent = old[: len(old) - len(old.lstrip())]
+        trailing = "," if old.rstrip().endswith(",") else ""
+        context = lines[model_idx - 1] if model_idx else ""
+        out += [
+            ("hunk", f"@@ -{model_idx},2 +{model_idx},2 @@"),
+            ("ctx", f" {context}"),
+            ("del", f"-{old}"),
+            ("add", f"+{indent}model=os.environ.get(\"NABLA_MODEL_{env_suffix}\", "
+                    f"\"{deployment['model']}\"){trailing}"),
+        ]
+
+    # Hunk 3: credentials, appended to the repo's .env.
+    out += [
+        ("head", "--- a/.env"),
+        ("head", "+++ b/.env"),
+        ("add", f"+{deployment['api_key_env']}=fsk-................ (provisioned by backend)"),
+        ("add", f"+NABLA_BASE_URL={deployment['base_url']}"),
+        ("add", f"+NABLA_MODEL_{env_suffix}={deployment['model']}"),
+    ]
+    return out
+
+
+def pace(seconds: float, interactive: bool) -> None:
+    """Small dramatic pause, interactive terminals only."""
+    if interactive:
+        time.sleep(seconds)
+
+
+class FillerSite:
+    """Stand-in site so single-site repos still demo a multi-row pull."""
+
+    symbol = "extract_invoice"
+    file = "app/enrichment.py"
+    span = (41, 58)
+
+
+def filler_deployment() -> tuple[FillerSite, dict]:
+    site = FillerSite()
+    dep = mock_deployment(site.symbol, "gpt-4o")
+    return site, dep
+
+
+def filler_diff(dep: dict) -> list[tuple[str, str]]:
+    """Canned diff for the filler site (its file doesn't exist on disk)."""
+    return [
+        ("head", f"--- a/{FillerSite.file}"),
+        ("head", f"+++ b/{FillerSite.file}"),
+        ("hunk", "@@ -44,2 +44,2 @@"),
+        ("ctx", "     response = client.chat.completions.create("),
+        ("del", '-        model="gpt-4o",'),
+        ("add", f'+        model=os.environ.get("NABLA_MODEL_EXTRACT_INVOICE", "{dep["model"]}"),'),
+        ("head", "--- a/.env"),
+        ("head", "+++ b/.env"),
+        ("add", f"+NABLA_MODEL_EXTRACT_INVOICE={dep['model']}"),
+    ]

{nabla_cli-0.9.0 → nabla_cli-0.9.2}/nabla/pusher.py

@@ -50,9 +50,15 @@ def resolve_endpoint(flag_value: str | None = None) -> str:
 
 def to_run_task(kind: str, artifact: dict) -> dict:
     """Translate a build/jeremy artifact into the dispatcher's
-    RunTaskRequest shape. Pure rearrangement — nothing invented."""
+    RunTaskRequest shape. Pure rearrangement — nothing invented.
+
+    site_id (the profile/seed's stable template+schema hash) rides along so
+    the backend can key the eventual fine-tuned deployment to the exact
+    call site — `nabla pull` joins on it later. Never file:line, never the
+    bare symbol: both are ambiguous across drift and duplicates."""
     if kind == "jeremy":
         return {
+            "site_id": artifact.get("site_id"),
             "goal": artifact.get("goal", ""),
             "examples": artifact.get("examples", []),
             "student_system_prompt": artifact.get("system_prompt", "") or "",
@@ -65,6 +71,7 @@ def to_run_task(kind: str, artifact: dict) -> dict:
         + [str(c).rstrip(".") + "." for c in goal_block.get("constraints", [])]
     ).strip()
     return {
+        "site_id": artifact.get("site_id"),
         "goal": goal_text,
         "examples": [
             {"problem": e.get("prompt", ""), "solution": e.get("completion", "")}

{nabla_cli-0.9.0 → nabla_cli-0.9.2}/nabla_cli.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nabla-cli
-Version: 0.9.0
+Version: 0.9.2
 Summary: Call-site harvester: scan OpenAI call sites, record real traffic, emit site profiles
 License: MIT
 Requires-Python: >=3.11

{nabla_cli-0.9.0 → nabla_cli-0.9.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "nabla-cli"
-version = "0.9.0"
+version = "0.9.2"
 description = "Call-site harvester: scan OpenAI call sites, record real traffic, emit site profiles"
 readme = "README.md"
 license = { text = "MIT" }

{nabla_cli-0.9.0 → nabla_cli-0.9.2}/tests/test_cli_ux.py

@@ -547,7 +547,9 @@ def test_jeremy_single_site_writes_seed_spec(tmp_path, monkeypatch, capsys):
 
     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 set(seed) == {"site_id", "symbol", "goal", "system_prompt", "examples"}
+    assert len(seed["site_id"]) == 64  # matches the profile's stable hash
+    assert seed["symbol"] == "extract_receipt"
     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"]

{nabla_cli-0.9.0 → nabla_cli-0.9.2}/tests/test_pull.py

@@ -1,76 +1,79 @@
-"""Tests for the nabla pull simulation — its one hard guarantee: it never
-writes anything."""
-
-from __future__ import annotations
-
-import json
-
-import pytest
-
-from nabla.cli import main
-
-from test_cli_ux import _SINGLE_SITE_SRC
-
-
-@pytest.fixture(autouse=True)
-def _clean_env(monkeypatch, tmp_path):
-    monkeypatch.setattr("nabla.config.config_path", lambda: tmp_path / "no-config.json")
-    monkeypatch.delenv("NABLA_PULL_ENDPOINT", raising=False)
-
-
-def _snapshot(root):
-    return {p: p.read_bytes() for p in root.rglob("*") if p.is_file()}
-
-
-def test_pull_dry_run_prints_diff_and_writes_nothing(tmp_path, monkeypatch, capsys):
-    repo = tmp_path / "repo"
-    repo.mkdir()
-    (repo / "single_call_site.py").write_text(_SINGLE_SITE_SRC, encoding="utf-8")
-    monkeypatch.chdir(tmp_path)
-    before = _snapshot(tmp_path)
-
-    rc = main(["pull", str(repo)])
-
-    assert rc == 0
-    assert _snapshot(tmp_path) == before  # nothing created or modified
-    out = capsys.readouterr().out
-    assert "freesolo/qwen3.5-0.8b-ft-extract-receipt" in out
-    assert "gpt-4o-mini" in out          # REPLACES column from real sampling
-    assert "--- a/single_call_site.py" in out
-    assert "+++ b/.env" in out
-    assert "dry run" in out
-    assert "simulation — nothing was written" in out
-
-
-def test_pull_apply_simulates_and_writes_nothing(tmp_path, monkeypatch, capsys):
-    repo = tmp_path / "repo"
-    repo.mkdir()
-    (repo / "single_call_site.py").write_text(_SINGLE_SITE_SRC, encoding="utf-8")
-    (tmp_path / "extract_receipt.profile.json").write_text(json.dumps(
-        {"baseline": {"cost_per_call_usd": 0.011, "p50_latency_ms": 2000,
-                      "calls_observed": 5}}), encoding="utf-8")
-    monkeypatch.chdir(tmp_path)
-    before = _snapshot(tmp_path)
-
-    rc = main(["pull", str(repo), "--apply"])
-
-    assert rc == 0
-    assert _snapshot(tmp_path) == before
-    out = capsys.readouterr().out
-    assert "2 edits -> single_call_site.py" in out
-    assert "completions valid" in out
-    assert "$0.011000 -> $0.001000" in out
-    assert "rollback: unset NABLA_MODEL_EXTRACT_RECEIPT" in out
-    assert "simulation — nothing was written" in out
-
-
-def test_pull_unknown_site_exits_2(tmp_path, monkeypatch, capsys):
-    repo = tmp_path / "repo"
-    repo.mkdir()
-    (repo / "single_call_site.py").write_text(_SINGLE_SITE_SRC, encoding="utf-8")
-    monkeypatch.chdir(tmp_path)
-
-    rc = main(["pull", str(repo), "--site", "nope"])
-
-    assert rc == 2
-    assert "no supported site" in capsys.readouterr().err
+"""Tests for the nabla pull simulation — its one hard guarantee: it never
+writes anything."""
+
+from __future__ import annotations
+
+import json
+
+import pytest
+
+from nabla.cli import main
+
+from test_cli_ux import _SINGLE_SITE_SRC
+
+
+@pytest.fixture(autouse=True)
+def _clean_env(monkeypatch, tmp_path):
+    monkeypatch.setattr("nabla.config.config_path", lambda: tmp_path / "no-config.json")
+    monkeypatch.delenv("NABLA_PULL_ENDPOINT", raising=False)
+
+
+def _snapshot(root):
+    return {p: p.read_bytes() for p in root.rglob("*") if p.is_file()}
+
+
+def test_pull_dry_run_prints_diff_and_writes_nothing(tmp_path, monkeypatch, capsys):
+    repo = tmp_path / "repo"
+    repo.mkdir()
+    (repo / "single_call_site.py").write_text(_SINGLE_SITE_SRC, encoding="utf-8")
+    monkeypatch.chdir(tmp_path)
+    before = _snapshot(tmp_path)
+
+    rc = main(["pull", str(repo)])
+
+    assert rc == 0
+    assert _snapshot(tmp_path) == before  # nothing created or modified
+    out = capsys.readouterr().out
+    assert "freesolo/qwen3.5-0.8b-ft-extract-receipt" in out
+    assert "gpt-4o-mini" in out          # REPLACES column from real sampling
+    assert "--- a/single_call_site.py" in out
+    assert "+++ b/.env" in out
+    assert "dry run" in out
+    assert "simulation — nothing was written" in out
+    # single-site repos are padded with a canned second deployment
+    assert "extract_invoice" in out
+    assert "freesolo/qwen3.5-0.8b-ft-extract-invoice" in out
+
+
+def test_pull_apply_simulates_and_writes_nothing(tmp_path, monkeypatch, capsys):
+    repo = tmp_path / "repo"
+    repo.mkdir()
+    (repo / "single_call_site.py").write_text(_SINGLE_SITE_SRC, encoding="utf-8")
+    (tmp_path / "extract_receipt.profile.json").write_text(json.dumps(
+        {"baseline": {"cost_per_call_usd": 0.011, "p50_latency_ms": 2000,
+                      "calls_observed": 5}}), encoding="utf-8")
+    monkeypatch.chdir(tmp_path)
+    before = _snapshot(tmp_path)
+
+    rc = main(["pull", str(repo), "--apply"])
+
+    assert rc == 0
+    assert _snapshot(tmp_path) == before
+    out = capsys.readouterr().out
+    assert "2 edits -> single_call_site.py" in out
+    assert "completions valid" in out
+    assert "$0.011000 -> $0.001000" in out
+    assert "rollback: unset NABLA_MODEL_EXTRACT_RECEIPT" in out
+    assert "simulation — nothing was written" in out
+
+
+def test_pull_unknown_site_exits_2(tmp_path, monkeypatch, capsys):
+    repo = tmp_path / "repo"
+    repo.mkdir()
+    (repo / "single_call_site.py").write_text(_SINGLE_SITE_SRC, encoding="utf-8")
+    monkeypatch.chdir(tmp_path)
+
+    rc = main(["pull", str(repo), "--site", "nope"])
+
+    assert rc == 2
+    assert "no supported site" in capsys.readouterr().err

{nabla_cli-0.9.0 → nabla_cli-0.9.2}/tests/test_push.py

@@ -70,6 +70,7 @@ def test_push_artifact_translates_jeremy_to_run_task():
         ok, detail = push_artifact(backend.url, "s", "jeremy", artifact)
         assert ok and "200" in detail
         assert backend.received == [{
+            "site_id": None,
             "goal": "Solve problems.",
             "examples": [{"problem": "p", "solution": "s"}],
             "student_system_prompt": "You are a tutor.",
@@ -114,7 +115,8 @@ def test_cmd_push_sweeps_site_artifacts(tmp_path, monkeypatch, capsys):
 
     assert rc == 0
     assert len(backend.received) == 2
-    assert all(set(r) == {"goal", "examples", "student_system_prompt", "student_goal"}
+    assert all(set(r) == {"site_id", "goal", "examples",
+                          "student_system_prompt", "student_goal"}
                for r in backend.received)
     assert "pushed 2 artifact(s)" in capsys.readouterr().out