shipwright-kit 0.10.0__tar.gz → 0.11.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {shipwright_kit-0.10.0/shipwright_kit.egg-info → shipwright_kit-0.11.0}/PKG-INFO +1 -1
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/pyproject.toml +1 -1
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/__init__.py +1 -1
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/llm.py +46 -59
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0/shipwright_kit.egg-info}/PKG-INFO +1 -1
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/tests/test_llm.py +150 -144
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/LICENSE +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/README.md +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/setup.cfg +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/cli.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/config.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/design/__init__.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/design/banner.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/design/console.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/design/glyphs.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/design/output.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/design/palette.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/design/tiers.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/eval/__init__.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/eval/corpus.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/eval/harness.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/eval/metrics.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/py.typed +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/security/__init__.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/security/eval.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/security/injection.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/security/ssrf.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit/security/theme.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit.egg-info/SOURCES.txt +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit.egg-info/dependency_links.txt +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit.egg-info/entry_points.txt +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit.egg-info/requires.txt +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit.egg-info/top_level.txt +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/tests/test_cli.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/tests/test_config.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/tests/test_packaging.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/tests/test_packs_entrypoint.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/tests/test_release_config.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/tests/test_template_wiring.py +0 -0
- {shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/tests/test_tooling.py +0 -0
|
@@ -7,7 +7,7 @@ build-backend = "setuptools.build_meta"
|
|
|
7
7
|
# PyPI distribution name. The bare `shipwright` is taken on PyPI (unrelated 6si
|
|
8
8
|
# tool), so the dist is `shipwright-kit`; the IMPORT name is `shipwright_kit`.
|
|
9
9
|
name = "shipwright-kit"
|
|
10
|
-
version = "0.
|
|
10
|
+
version = "0.11.0"
|
|
11
11
|
description = "Shipwright — AI-agent dev framework + import-light design/eval/security library"
|
|
12
12
|
readme = "README.md"
|
|
13
13
|
requires-python = ">=3.11"
|
|
@@ -4,8 +4,16 @@ Owns the *mechanism* only — "build request -> call SDK/HTTP -> return raw
|
|
|
4
4
|
text" for Anthropic, OpenAI, and a local Ollama server. Mirrors
|
|
5
5
|
``shipwright_kit.config``'s shape: shared mechanism, per-tool schema. The
|
|
6
6
|
per-tool bits (prompt content, response JSON-parse/validate/fence-strip,
|
|
7
|
-
template fallback, the ``SummarizerProtocol``/``ExplainerProtocol``,
|
|
8
|
-
try/except around these calls) stay
|
|
7
|
+
template fallback, the ``SummarizerProtocol``/``ExplainerProtocol``, client
|
|
8
|
+
construction/import-guarding, and every try/except around these calls) stay
|
|
9
|
+
in each tool.
|
|
10
|
+
|
|
11
|
+
The SDK-backed providers (``anthropic_complete``, ``openai_complete``) take
|
|
12
|
+
an INJECTED ``client``: the caller constructs its own SDK client (with its
|
|
13
|
+
own install-hint/``ImportError`` handling in the tool's ``__init__``) and
|
|
14
|
+
passes it in. This module never imports ``anthropic``/``openai`` and never
|
|
15
|
+
constructs a client itself — the client is duck-typed, so import-lightness
|
|
16
|
+
here is structural (no SDK import exists to lazily defer), not merely lazy.
|
|
9
17
|
|
|
10
18
|
These are DUMB TRANSPORTS. They do NOT redact, sanitize, or scan ``system``/
|
|
11
19
|
``user`` text in any way — whatever the caller passes is sent to the provider
|
|
@@ -16,11 +24,9 @@ Exception-transparent by design: none of the three public functions contains
|
|
|
16
24
|
a ``try``/``except``. SDK, HTTP, and JSON/KeyError failures propagate to the
|
|
17
25
|
caller unchanged so each tool can keep its own existing error handling
|
|
18
26
|
(re-raise as ``RuntimeError``, swallow-and-fall-back-to-template, etc.) without
|
|
19
|
-
this module making that policy choice for them. The only
|
|
20
|
-
*by* this module itself
|
|
21
|
-
``system_mode`` literal
|
|
22
|
-
tool-specific install hint) when the ``anthropic``/``openai`` packages are not
|
|
23
|
-
installed — both are input-validation guards, not error-swallowing.
|
|
27
|
+
this module making that policy choice for them. The only exception raised
|
|
28
|
+
*by* this module itself is ``ValueError`` for an unrecognized ``extract``/
|
|
29
|
+
``system_mode`` literal — an input-validation guard, not error-swallowing.
|
|
24
30
|
|
|
25
31
|
No ``max_tokens``/``temperature`` default is baked in here (the config.py
|
|
26
32
|
lesson: no schema in the mechanism) — callers supply every value.
|
|
@@ -38,29 +44,6 @@ from typing import Any
|
|
|
38
44
|
__all__ = ["anthropic_complete", "ollama_generate", "openai_complete"]
|
|
39
45
|
|
|
40
46
|
|
|
41
|
-
# ---------------------------------------------------------------------------
|
|
42
|
-
# Lazy SDK imports — kept out of the 3 public functions so those stay
|
|
43
|
-
# try/except-free; `import shipwright_kit.llm` itself never touches
|
|
44
|
-
# anthropic/openai and stays stdlib-only.
|
|
45
|
-
# ---------------------------------------------------------------------------
|
|
46
|
-
|
|
47
|
-
|
|
48
|
-
def _import_anthropic(install_hint: str) -> Any:
|
|
49
|
-
try:
|
|
50
|
-
import anthropic # noqa: PLC0415
|
|
51
|
-
except ImportError as exc:
|
|
52
|
-
raise ImportError(install_hint) from exc
|
|
53
|
-
return anthropic
|
|
54
|
-
|
|
55
|
-
|
|
56
|
-
def _import_openai(install_hint: str) -> Any:
|
|
57
|
-
try:
|
|
58
|
-
import openai # noqa: PLC0415
|
|
59
|
-
except ImportError as exc:
|
|
60
|
-
raise ImportError(install_hint) from exc
|
|
61
|
-
return openai
|
|
62
|
-
|
|
63
|
-
|
|
64
47
|
# ---------------------------------------------------------------------------
|
|
65
48
|
# Anthropic
|
|
66
49
|
# ---------------------------------------------------------------------------
|
|
@@ -68,19 +51,21 @@ def _import_openai(install_hint: str) -> Any:
|
|
|
68
51
|
|
|
69
52
|
def anthropic_complete(
|
|
70
53
|
*,
|
|
71
|
-
|
|
54
|
+
client: Any,
|
|
72
55
|
model: str,
|
|
73
56
|
max_tokens: int,
|
|
74
57
|
system: str,
|
|
75
58
|
user: str,
|
|
76
|
-
install_hint: str,
|
|
77
59
|
temperature: float | None = None,
|
|
78
60
|
extract: str = "first_text_block",
|
|
79
61
|
) -> str:
|
|
80
|
-
"""Call the Anthropic Messages API
|
|
62
|
+
"""Call the Anthropic Messages API on a caller-supplied client and return
|
|
63
|
+
the extracted text.
|
|
81
64
|
|
|
82
|
-
|
|
83
|
-
|
|
65
|
+
``client`` is a pre-built Anthropic SDK client (e.g.
|
|
66
|
+
``anthropic.Anthropic(api_key=...)``) constructed and owned by the
|
|
67
|
+
caller; this function only calls ``client.messages.create(...)`` and
|
|
68
|
+
extracts text from the response. ``extract`` reconciles the two known
|
|
84
69
|
current extraction behaviors:
|
|
85
70
|
|
|
86
71
|
- ``"first_text_block"`` (sift's current behavior): scan
|
|
@@ -92,17 +77,10 @@ def anthropic_complete(
|
|
|
92
77
|
behavior is preserved on purpose (named follow-up F2 owns fixing it).
|
|
93
78
|
|
|
94
79
|
Raises:
|
|
95
|
-
ImportError: the ``anthropic`` package is not installed; the message
|
|
96
|
-
is exactly ``install_hint`` (each tool supplies its own text so
|
|
97
|
-
this function never bakes in tool-specific wording).
|
|
98
80
|
ValueError: ``extract`` is not one of the two known modes.
|
|
99
|
-
Exception: any exception raised by ``
|
|
100
|
-
|
|
101
|
-
here.
|
|
81
|
+
Exception: any exception raised by ``client.messages.create(...)``
|
|
82
|
+
propagates unchanged — no try/except here.
|
|
102
83
|
"""
|
|
103
|
-
anthropic = _import_anthropic(install_hint)
|
|
104
|
-
client = anthropic.Anthropic(api_key=api_key)
|
|
105
|
-
|
|
106
84
|
kwargs: dict[str, Any] = {
|
|
107
85
|
"model": model,
|
|
108
86
|
"max_tokens": max_tokens,
|
|
@@ -131,31 +109,28 @@ def anthropic_complete(
|
|
|
131
109
|
|
|
132
110
|
def openai_complete(
|
|
133
111
|
*,
|
|
134
|
-
|
|
112
|
+
client: Any,
|
|
135
113
|
model: str,
|
|
136
114
|
max_tokens: int,
|
|
137
115
|
system: str,
|
|
138
116
|
user: str,
|
|
139
|
-
install_hint: str,
|
|
140
117
|
temperature: float | None = None,
|
|
141
118
|
) -> str:
|
|
142
|
-
"""Call the OpenAI Chat Completions API
|
|
119
|
+
"""Call the OpenAI Chat Completions API on a caller-supplied client and
|
|
120
|
+
return the response text.
|
|
143
121
|
|
|
144
|
-
|
|
145
|
-
|
|
146
|
-
|
|
147
|
-
|
|
122
|
+
``client`` is a pre-built OpenAI SDK client (e.g.
|
|
123
|
+
``openai.OpenAI(api_key=...)``) constructed and owned by the caller; this
|
|
124
|
+
function only calls ``client.chat.completions.create(...)`` and extracts
|
|
125
|
+
text from the response (``response.choices[0].message.content or ""`` —
|
|
126
|
+
a ``None`` content, e.g. a tool-call-only response, degrades to ``""``
|
|
127
|
+
rather than raising, matching both sift and barb today).
|
|
148
128
|
|
|
149
129
|
Raises:
|
|
150
|
-
|
|
151
|
-
|
|
152
|
-
|
|
153
|
-
``.chat.completions.create(...)`` propagates unchanged — no
|
|
154
|
-
try/except here.
|
|
130
|
+
Exception: any exception raised by
|
|
131
|
+
``client.chat.completions.create(...)`` propagates unchanged —
|
|
132
|
+
no try/except here.
|
|
155
133
|
"""
|
|
156
|
-
openai = _import_openai(install_hint)
|
|
157
|
-
client = openai.OpenAI(api_key=api_key)
|
|
158
|
-
|
|
159
134
|
kwargs: dict[str, Any] = {
|
|
160
135
|
"model": model,
|
|
161
136
|
"max_tokens": max_tokens,
|
|
@@ -184,6 +159,7 @@ def ollama_generate(
|
|
|
184
159
|
user: str,
|
|
185
160
|
timeout: float | None,
|
|
186
161
|
system_mode: str,
|
|
162
|
+
options: dict[str, Any] | None = None,
|
|
187
163
|
) -> str:
|
|
188
164
|
"""POST to a local Ollama ``/api/generate`` endpoint and return the raw
|
|
189
165
|
``response`` field, unstripped.
|
|
@@ -200,6 +176,14 @@ def ollama_generate(
|
|
|
200
176
|
top-level payload key, ``prompt`` is ``user`` alone. Payload keys:
|
|
201
177
|
``{"model", "system", "prompt", "stream"}``.
|
|
202
178
|
|
|
179
|
+
``options`` is an ADDITIVE, OPTIONAL passthrough for Ollama's
|
|
180
|
+
per-request generation options (vex's current behavior, e.g.
|
|
181
|
+
``{"num_predict": ..., "temperature": ...}``). When given (not
|
|
182
|
+
``None``), it is included verbatim as the ``"options"`` payload key.
|
|
183
|
+
When omitted (the default, ``None``), no ``"options"`` key is added and
|
|
184
|
+
the payload is byte-identical to today — sift and barb, which never
|
|
185
|
+
pass this argument, are unaffected.
|
|
186
|
+
|
|
203
187
|
Note: barb's caller additionally does ``.strip()`` on the returned text
|
|
204
188
|
today; that is NOT done here (sift's raw extraction never strips) — each
|
|
205
189
|
tool's own call site is responsible for any such post-processing.
|
|
@@ -222,6 +206,9 @@ def ollama_generate(
|
|
|
222
206
|
else:
|
|
223
207
|
raise ValueError(f"unknown system_mode: {system_mode!r}")
|
|
224
208
|
|
|
209
|
+
if options is not None:
|
|
210
|
+
payload["options"] = options
|
|
211
|
+
|
|
225
212
|
data = json.dumps(payload).encode("utf-8")
|
|
226
213
|
req = urllib.request.Request(
|
|
227
214
|
generate_url,
|
|
@@ -4,8 +4,10 @@ Pins the exact request shapes both sift's characterization tests
|
|
|
4
4
|
(``sift/tests/test_llm_provider_requests.py``) and barb's
|
|
5
5
|
(``barb/tests/test_explain_llm_providers.py``) expect, so that when each tool
|
|
6
6
|
is retrofitted onto this module in a later phase, those suites pass
|
|
7
|
-
byte-identical. All
|
|
8
|
-
real API keys
|
|
7
|
+
byte-identical. All clients are plain mocks passed in directly (client
|
|
8
|
+
INJECTION, not internal construction) — no live network, no real API keys,
|
|
9
|
+
and no faking of ``sys.modules['anthropic']``/``['openai']`` since this
|
|
10
|
+
module no longer imports either SDK.
|
|
9
11
|
"""
|
|
10
12
|
|
|
11
13
|
from __future__ import annotations
|
|
@@ -14,7 +16,6 @@ import json
|
|
|
14
16
|
import subprocess
|
|
15
17
|
import sys
|
|
16
18
|
import urllib.error
|
|
17
|
-
from types import SimpleNamespace
|
|
18
19
|
from unittest.mock import MagicMock
|
|
19
20
|
|
|
20
21
|
import pytest
|
|
@@ -26,31 +27,25 @@ from shipwright_kit.llm import anthropic_complete, ollama_generate, openai_compl
|
|
|
26
27
|
# ---------------------------------------------------------------------------
|
|
27
28
|
|
|
28
29
|
|
|
29
|
-
def
|
|
30
|
-
"""
|
|
31
|
-
|
|
32
|
-
so tests can assert on ``Anthropic(...)`` / ``.messages.create(...)``."""
|
|
30
|
+
def _make_anthropic_client(response_text: str = "ok") -> MagicMock:
|
|
31
|
+
"""Build a mock Anthropic SDK client (what a tool's ``__init__`` would
|
|
32
|
+
construct and inject as ``self._client``)."""
|
|
33
33
|
mock_client = MagicMock()
|
|
34
34
|
mock_response = MagicMock()
|
|
35
35
|
mock_response.content = [MagicMock(text=response_text)]
|
|
36
36
|
mock_client.messages.create.return_value = mock_response
|
|
37
|
-
|
|
38
|
-
mock_anthropic_cls = MagicMock(return_value=mock_client)
|
|
39
|
-
fake_module = SimpleNamespace(Anthropic=mock_anthropic_cls)
|
|
40
|
-
monkeypatch.setitem(sys.modules, "anthropic", fake_module)
|
|
41
|
-
return mock_client, mock_anthropic_cls
|
|
37
|
+
return mock_client
|
|
42
38
|
|
|
43
39
|
|
|
44
40
|
class TestAnthropicRequestConstruction:
|
|
45
|
-
def test_sends_model_max_tokens_system_and_user_message(self
|
|
46
|
-
mock_client
|
|
41
|
+
def test_sends_model_max_tokens_system_and_user_message(self):
|
|
42
|
+
mock_client = _make_anthropic_client()
|
|
47
43
|
anthropic_complete(
|
|
48
|
-
|
|
44
|
+
client=mock_client,
|
|
49
45
|
model="claude-sonnet-4-6",
|
|
50
46
|
max_tokens=2048,
|
|
51
47
|
system="sys prompt",
|
|
52
48
|
user="user prompt",
|
|
53
|
-
install_hint="pip install x[llm]",
|
|
54
49
|
)
|
|
55
50
|
kwargs = mock_client.messages.create.call_args.kwargs
|
|
56
51
|
assert kwargs["model"] == "claude-sonnet-4-6"
|
|
@@ -58,54 +53,29 @@ class TestAnthropicRequestConstruction:
|
|
|
58
53
|
assert kwargs["system"] == "sys prompt"
|
|
59
54
|
assert kwargs["messages"] == [{"role": "user", "content": "user prompt"}]
|
|
60
55
|
|
|
61
|
-
def test_temperature_omitted_when_none(self
|
|
62
|
-
mock_client
|
|
63
|
-
anthropic_complete(
|
|
64
|
-
api_key="fake-key",
|
|
65
|
-
model="m",
|
|
66
|
-
max_tokens=10,
|
|
67
|
-
system="s",
|
|
68
|
-
user="u",
|
|
69
|
-
install_hint="hint",
|
|
70
|
-
)
|
|
56
|
+
def test_temperature_omitted_when_none(self):
|
|
57
|
+
mock_client = _make_anthropic_client()
|
|
58
|
+
anthropic_complete(client=mock_client, model="m", max_tokens=10, system="s", user="u")
|
|
71
59
|
assert "temperature" not in mock_client.messages.create.call_args.kwargs
|
|
72
60
|
|
|
73
|
-
def test_temperature_included_when_given(self
|
|
74
|
-
mock_client
|
|
75
|
-
anthropic_complete(
|
|
76
|
-
api_key="fake-key",
|
|
77
|
-
model="m",
|
|
78
|
-
max_tokens=10,
|
|
79
|
-
system="s",
|
|
80
|
-
user="u",
|
|
81
|
-
install_hint="hint",
|
|
82
|
-
temperature=0.42,
|
|
83
|
-
)
|
|
61
|
+
def test_temperature_included_when_given(self):
|
|
62
|
+
mock_client = _make_anthropic_client()
|
|
63
|
+
anthropic_complete(client=mock_client, model="m", max_tokens=10, system="s", user="u", temperature=0.42)
|
|
84
64
|
assert mock_client.messages.create.call_args.kwargs["temperature"] == 0.42
|
|
85
65
|
|
|
86
|
-
def test_no_response_format_or_tools_param_sent(self
|
|
87
|
-
mock_client
|
|
88
|
-
anthropic_complete(
|
|
66
|
+
def test_no_response_format_or_tools_param_sent(self):
|
|
67
|
+
mock_client = _make_anthropic_client()
|
|
68
|
+
anthropic_complete(client=mock_client, model="m", max_tokens=10, system="s", user="u")
|
|
89
69
|
kwargs = mock_client.messages.create.call_args.kwargs
|
|
90
70
|
assert "response_format" not in kwargs
|
|
91
71
|
assert "tools" not in kwargs
|
|
92
72
|
|
|
93
|
-
def
|
|
94
|
-
|
|
95
|
-
|
|
96
|
-
|
|
97
|
-
|
|
98
|
-
|
|
99
|
-
monkeypatch.setitem(sys.modules, "anthropic", None) # forces ImportError on `import anthropic`
|
|
100
|
-
with pytest.raises(ImportError, match="pip install sift-triage\\[llm\\]"):
|
|
101
|
-
anthropic_complete(
|
|
102
|
-
api_key="k",
|
|
103
|
-
model="m",
|
|
104
|
-
max_tokens=10,
|
|
105
|
-
system="s",
|
|
106
|
-
user="u",
|
|
107
|
-
install_hint="pip install sift-triage[llm]",
|
|
108
|
-
)
|
|
73
|
+
def test_uses_the_injected_client_not_a_new_one(self):
|
|
74
|
+
"""No internal client construction: the exact object passed in is the
|
|
75
|
+
one whose .messages.create(...) gets called."""
|
|
76
|
+
mock_client = _make_anthropic_client()
|
|
77
|
+
anthropic_complete(client=mock_client, model="m", max_tokens=10, system="s", user="u")
|
|
78
|
+
mock_client.messages.create.assert_called_once()
|
|
109
79
|
|
|
110
80
|
|
|
111
81
|
# ---------------------------------------------------------------------------
|
|
@@ -114,64 +84,58 @@ class TestAnthropicRequestConstruction:
|
|
|
114
84
|
|
|
115
85
|
|
|
116
86
|
class TestAnthropicExtraction:
|
|
117
|
-
def test_first_text_block_skips_leading_non_text_block(self
|
|
118
|
-
mock_client
|
|
87
|
+
def test_first_text_block_skips_leading_non_text_block(self):
|
|
88
|
+
mock_client = _make_anthropic_client()
|
|
119
89
|
|
|
120
90
|
class NoTextBlock:
|
|
121
91
|
type = "tool_use"
|
|
122
92
|
|
|
123
93
|
mock_client.messages.create.return_value.content = [NoTextBlock(), MagicMock(text="the answer")]
|
|
124
94
|
result = anthropic_complete(
|
|
125
|
-
|
|
95
|
+
client=mock_client, model="m", max_tokens=10, system="s", user="u", extract="first_text_block"
|
|
126
96
|
)
|
|
127
97
|
assert result == "the answer"
|
|
128
98
|
|
|
129
|
-
def test_first_text_block_returns_empty_string_when_no_text_block(self
|
|
130
|
-
mock_client
|
|
99
|
+
def test_first_text_block_returns_empty_string_when_no_text_block(self):
|
|
100
|
+
mock_client = _make_anthropic_client()
|
|
131
101
|
|
|
132
102
|
class NoTextBlock:
|
|
133
103
|
type = "tool_use"
|
|
134
104
|
|
|
135
105
|
mock_client.messages.create.return_value.content = [NoTextBlock()]
|
|
136
106
|
result = anthropic_complete(
|
|
137
|
-
|
|
107
|
+
client=mock_client, model="m", max_tokens=10, system="s", user="u", extract="first_text_block"
|
|
138
108
|
)
|
|
139
109
|
assert result == ""
|
|
140
110
|
|
|
141
|
-
def test_index0_returns_content_0_text(self
|
|
142
|
-
mock_client
|
|
111
|
+
def test_index0_returns_content_0_text(self):
|
|
112
|
+
mock_client = _make_anthropic_client(response_text="direct")
|
|
143
113
|
result = anthropic_complete(
|
|
144
|
-
|
|
114
|
+
client=mock_client, model="m", max_tokens=10, system="s", user="u", extract="index0"
|
|
145
115
|
)
|
|
146
116
|
assert result == "direct"
|
|
147
117
|
|
|
148
|
-
def test_index0_raises_index_error_on_empty_content_uncaught(self
|
|
118
|
+
def test_index0_raises_index_error_on_empty_content_uncaught(self):
|
|
149
119
|
"""Preserves barb's current crash-on-empty; NOT fixed here (follow-up F2)."""
|
|
150
|
-
mock_client
|
|
120
|
+
mock_client = _make_anthropic_client()
|
|
151
121
|
mock_client.messages.create.return_value.content = []
|
|
152
122
|
with pytest.raises(IndexError):
|
|
153
|
-
anthropic_complete(
|
|
154
|
-
api_key="k", model="m", max_tokens=10, system="s", user="u", install_hint="h", extract="index0"
|
|
155
|
-
)
|
|
123
|
+
anthropic_complete(client=mock_client, model="m", max_tokens=10, system="s", user="u", extract="index0")
|
|
156
124
|
|
|
157
|
-
def test_index0_raises_attribute_error_when_first_block_has_no_text(self
|
|
158
|
-
mock_client
|
|
125
|
+
def test_index0_raises_attribute_error_when_first_block_has_no_text(self):
|
|
126
|
+
mock_client = _make_anthropic_client()
|
|
159
127
|
|
|
160
128
|
class NoTextBlock:
|
|
161
129
|
type = "tool_use"
|
|
162
130
|
|
|
163
131
|
mock_client.messages.create.return_value.content = [NoTextBlock()]
|
|
164
132
|
with pytest.raises(AttributeError):
|
|
165
|
-
anthropic_complete(
|
|
166
|
-
api_key="k", model="m", max_tokens=10, system="s", user="u", install_hint="h", extract="index0"
|
|
167
|
-
)
|
|
133
|
+
anthropic_complete(client=mock_client, model="m", max_tokens=10, system="s", user="u", extract="index0")
|
|
168
134
|
|
|
169
|
-
def test_unknown_extract_mode_raises_value_error(self
|
|
170
|
-
|
|
135
|
+
def test_unknown_extract_mode_raises_value_error(self):
|
|
136
|
+
mock_client = _make_anthropic_client()
|
|
171
137
|
with pytest.raises(ValueError, match="unknown extract mode"):
|
|
172
|
-
anthropic_complete(
|
|
173
|
-
api_key="k", model="m", max_tokens=10, system="s", user="u", install_hint="h", extract="bogus"
|
|
174
|
-
)
|
|
138
|
+
anthropic_complete(client=mock_client, model="m", max_tokens=10, system="s", user="u", extract="bogus")
|
|
175
139
|
|
|
176
140
|
|
|
177
141
|
# ---------------------------------------------------------------------------
|
|
@@ -180,19 +144,24 @@ class TestAnthropicExtraction:
|
|
|
180
144
|
|
|
181
145
|
|
|
182
146
|
class TestAnthropicExceptionTransparency:
|
|
183
|
-
def test_sdk_error_from_create_propagates_uncaught(self
|
|
184
|
-
mock_client
|
|
147
|
+
def test_sdk_error_from_create_propagates_uncaught(self):
|
|
148
|
+
mock_client = _make_anthropic_client()
|
|
185
149
|
mock_client.messages.create.side_effect = RuntimeError("boom from SDK")
|
|
186
150
|
with pytest.raises(RuntimeError, match="boom from SDK"):
|
|
187
|
-
anthropic_complete(
|
|
188
|
-
|
|
189
|
-
def
|
|
190
|
-
|
|
191
|
-
|
|
192
|
-
|
|
151
|
+
anthropic_complete(client=mock_client, model="m", max_tokens=10, system="s", user="u")
|
|
152
|
+
|
|
153
|
+
def test_client_raised_exception_propagates_without_wrapping(self):
|
|
154
|
+
"""llm.py no longer takes an api_key (client construction/credentials
|
|
155
|
+
are entirely the caller's concern), so there is nothing for this
|
|
156
|
+
function to leak into an exception message. What's left to pin: a
|
|
157
|
+
client-raised exception propagates completely unmodified — same
|
|
158
|
+
type, same message, no wrapping/added context from this module."""
|
|
159
|
+
mock_client = _make_anthropic_client()
|
|
160
|
+
original = RuntimeError("rate limited")
|
|
161
|
+
mock_client.messages.create.side_effect = original
|
|
193
162
|
with pytest.raises(RuntimeError) as excinfo:
|
|
194
|
-
anthropic_complete(
|
|
195
|
-
assert
|
|
163
|
+
anthropic_complete(client=mock_client, model="m", max_tokens=10, system="s", user="u")
|
|
164
|
+
assert excinfo.value is original
|
|
196
165
|
|
|
197
166
|
|
|
198
167
|
# ---------------------------------------------------------------------------
|
|
@@ -200,28 +169,23 @@ class TestAnthropicExceptionTransparency:
|
|
|
200
169
|
# ---------------------------------------------------------------------------
|
|
201
170
|
|
|
202
171
|
|
|
203
|
-
def
|
|
172
|
+
def _make_openai_client(content: str | None = "ok") -> MagicMock:
|
|
204
173
|
mock_client = MagicMock()
|
|
205
174
|
mock_response = MagicMock()
|
|
206
175
|
mock_response.choices = [MagicMock(message=MagicMock(content=content))]
|
|
207
176
|
mock_client.chat.completions.create.return_value = mock_response
|
|
208
|
-
|
|
209
|
-
mock_openai_cls = MagicMock(return_value=mock_client)
|
|
210
|
-
fake_module = SimpleNamespace(OpenAI=mock_openai_cls)
|
|
211
|
-
monkeypatch.setitem(sys.modules, "openai", fake_module)
|
|
212
|
-
return mock_client, mock_openai_cls
|
|
177
|
+
return mock_client
|
|
213
178
|
|
|
214
179
|
|
|
215
180
|
class TestOpenAIRequestConstruction:
|
|
216
|
-
def test_sends_model_max_tokens_and_two_role_messages(self
|
|
217
|
-
mock_client
|
|
181
|
+
def test_sends_model_max_tokens_and_two_role_messages(self):
|
|
182
|
+
mock_client = _make_openai_client()
|
|
218
183
|
openai_complete(
|
|
219
|
-
|
|
184
|
+
client=mock_client,
|
|
220
185
|
model="gpt-4o",
|
|
221
186
|
max_tokens=1024,
|
|
222
187
|
system="sys prompt",
|
|
223
188
|
user="user prompt",
|
|
224
|
-
install_hint="h",
|
|
225
189
|
)
|
|
226
190
|
kwargs = mock_client.chat.completions.create.call_args.kwargs
|
|
227
191
|
assert kwargs["model"] == "gpt-4o"
|
|
@@ -231,35 +195,28 @@ class TestOpenAIRequestConstruction:
|
|
|
231
195
|
{"role": "user", "content": "user prompt"},
|
|
232
196
|
]
|
|
233
197
|
|
|
234
|
-
def test_temperature_omitted_when_none(self
|
|
235
|
-
mock_client
|
|
236
|
-
openai_complete(
|
|
198
|
+
def test_temperature_omitted_when_none(self):
|
|
199
|
+
mock_client = _make_openai_client()
|
|
200
|
+
openai_complete(client=mock_client, model="m", max_tokens=10, system="s", user="u")
|
|
237
201
|
assert "temperature" not in mock_client.chat.completions.create.call_args.kwargs
|
|
238
202
|
|
|
239
|
-
def test_temperature_included_when_given(self
|
|
240
|
-
mock_client
|
|
241
|
-
openai_complete(
|
|
203
|
+
def test_temperature_included_when_given(self):
|
|
204
|
+
mock_client = _make_openai_client()
|
|
205
|
+
openai_complete(client=mock_client, model="m", max_tokens=10, system="s", user="u", temperature=0.1)
|
|
242
206
|
assert mock_client.chat.completions.create.call_args.kwargs["temperature"] == 0.1
|
|
243
207
|
|
|
244
|
-
def test_no_response_format_or_tools_param_sent(self
|
|
245
|
-
mock_client
|
|
246
|
-
openai_complete(
|
|
208
|
+
def test_no_response_format_or_tools_param_sent(self):
|
|
209
|
+
mock_client = _make_openai_client()
|
|
210
|
+
openai_complete(client=mock_client, model="m", max_tokens=10, system="s", user="u")
|
|
247
211
|
kwargs = mock_client.chat.completions.create.call_args.kwargs
|
|
248
212
|
assert "response_format" not in kwargs
|
|
249
213
|
assert "tools" not in kwargs
|
|
250
214
|
assert "functions" not in kwargs
|
|
251
215
|
|
|
252
|
-
def
|
|
253
|
-
|
|
254
|
-
openai_complete(
|
|
255
|
-
|
|
256
|
-
|
|
257
|
-
def test_import_error_uses_caller_supplied_install_hint(self, monkeypatch):
|
|
258
|
-
monkeypatch.setitem(sys.modules, "openai", None)
|
|
259
|
-
with pytest.raises(ImportError, match="pip install barb-phish\\[llm\\]"):
|
|
260
|
-
openai_complete(
|
|
261
|
-
api_key="k", model="m", max_tokens=10, system="s", user="u", install_hint="pip install barb-phish[llm]"
|
|
262
|
-
)
|
|
216
|
+
def test_uses_the_injected_client_not_a_new_one(self):
|
|
217
|
+
mock_client = _make_openai_client()
|
|
218
|
+
openai_complete(client=mock_client, model="m", max_tokens=10, system="s", user="u")
|
|
219
|
+
mock_client.chat.completions.create.assert_called_once()
|
|
263
220
|
|
|
264
221
|
|
|
265
222
|
# ---------------------------------------------------------------------------
|
|
@@ -268,14 +225,14 @@ class TestOpenAIRequestConstruction:
|
|
|
268
225
|
|
|
269
226
|
|
|
270
227
|
class TestOpenAIExtraction:
|
|
271
|
-
def test_returns_choices_0_message_content(self
|
|
272
|
-
|
|
273
|
-
result = openai_complete(
|
|
228
|
+
def test_returns_choices_0_message_content(self):
|
|
229
|
+
mock_client = _make_openai_client(content="direct answer")
|
|
230
|
+
result = openai_complete(client=mock_client, model="m", max_tokens=10, system="s", user="u")
|
|
274
231
|
assert result == "direct answer"
|
|
275
232
|
|
|
276
|
-
def test_none_content_degrades_to_empty_string(self
|
|
277
|
-
|
|
278
|
-
result = openai_complete(
|
|
233
|
+
def test_none_content_degrades_to_empty_string(self):
|
|
234
|
+
mock_client = _make_openai_client(content=None)
|
|
235
|
+
result = openai_complete(client=mock_client, model="m", max_tokens=10, system="s", user="u")
|
|
279
236
|
assert result == ""
|
|
280
237
|
|
|
281
238
|
|
|
@@ -285,19 +242,22 @@ class TestOpenAIExtraction:
|
|
|
285
242
|
|
|
286
243
|
|
|
287
244
|
class TestOpenAIExceptionTransparency:
|
|
288
|
-
def test_sdk_error_from_create_propagates_uncaught(self
|
|
289
|
-
mock_client
|
|
245
|
+
def test_sdk_error_from_create_propagates_uncaught(self):
|
|
246
|
+
mock_client = _make_openai_client()
|
|
290
247
|
mock_client.chat.completions.create.side_effect = RuntimeError("boom from SDK")
|
|
291
248
|
with pytest.raises(RuntimeError, match="boom from SDK"):
|
|
292
|
-
openai_complete(
|
|
293
|
-
|
|
294
|
-
def
|
|
295
|
-
|
|
296
|
-
|
|
297
|
-
|
|
249
|
+
openai_complete(client=mock_client, model="m", max_tokens=10, system="s", user="u")
|
|
250
|
+
|
|
251
|
+
def test_client_raised_exception_propagates_without_wrapping(self):
|
|
252
|
+
"""Mirrors the Anthropic case: no api_key in this API anymore, so the
|
|
253
|
+
pinned contract is that a client-raised exception passes through
|
|
254
|
+
this module completely unmodified."""
|
|
255
|
+
mock_client = _make_openai_client()
|
|
256
|
+
original = RuntimeError("rate limited")
|
|
257
|
+
mock_client.chat.completions.create.side_effect = original
|
|
298
258
|
with pytest.raises(RuntimeError) as excinfo:
|
|
299
|
-
openai_complete(
|
|
300
|
-
assert
|
|
259
|
+
openai_complete(client=mock_client, model="m", max_tokens=10, system="s", user="u")
|
|
260
|
+
assert excinfo.value is original
|
|
301
261
|
|
|
302
262
|
|
|
303
263
|
# ---------------------------------------------------------------------------
|
|
@@ -409,6 +369,50 @@ class TestOllamaRequestConstruction:
|
|
|
409
369
|
base_url="http://localhost:11434", model="m", system="s", user="u", timeout=None, system_mode="bogus"
|
|
410
370
|
)
|
|
411
371
|
|
|
372
|
+
def test_options_included_verbatim_when_given(self, monkeypatch):
|
|
373
|
+
"""vex's shape: a per-request options dict is passed through as-is."""
|
|
374
|
+
captured = _install_fake_urlopen(monkeypatch, {"response": "ok"})
|
|
375
|
+
opts = {"num_predict": 500, "temperature": 0.3}
|
|
376
|
+
ollama_generate(
|
|
377
|
+
base_url="http://localhost:11434",
|
|
378
|
+
model="m",
|
|
379
|
+
system="s",
|
|
380
|
+
user="u",
|
|
381
|
+
timeout=None,
|
|
382
|
+
system_mode="fold",
|
|
383
|
+
options=opts,
|
|
384
|
+
)
|
|
385
|
+
payload = captured["payload"]
|
|
386
|
+
assert payload["options"] == opts
|
|
387
|
+
|
|
388
|
+
def test_options_omitted_key_when_none(self, monkeypatch):
|
|
389
|
+
"""Regression guard: sift/barb never pass options, so with the default
|
|
390
|
+
(None) the payload must have no "options" key at all — byte-identical
|
|
391
|
+
to before this parameter existed."""
|
|
392
|
+
captured = _install_fake_urlopen(monkeypatch, {"response": "ok"})
|
|
393
|
+
ollama_generate(
|
|
394
|
+
base_url="http://localhost:11434", model="m", system="s", user="u", timeout=None, system_mode="fold"
|
|
395
|
+
)
|
|
396
|
+
assert "options" not in captured["payload"]
|
|
397
|
+
|
|
398
|
+
def test_options_combined_with_field_system_mode(self, monkeypatch):
|
|
399
|
+
"""vex's actual shape: system_mode="field" + options together produce
|
|
400
|
+
exactly {model, prompt, stream, options, system}."""
|
|
401
|
+
captured = _install_fake_urlopen(monkeypatch, {"response": "ok"})
|
|
402
|
+
opts = {"num_predict": 200, "temperature": 0.7}
|
|
403
|
+
ollama_generate(
|
|
404
|
+
base_url="http://localhost:11434",
|
|
405
|
+
model="m",
|
|
406
|
+
system="SYS",
|
|
407
|
+
user="USER",
|
|
408
|
+
timeout=None,
|
|
409
|
+
system_mode="field",
|
|
410
|
+
options=opts,
|
|
411
|
+
)
|
|
412
|
+
payload = captured["payload"]
|
|
413
|
+
assert set(payload.keys()) == {"model", "system", "prompt", "stream", "options"}
|
|
414
|
+
assert payload["options"] == opts
|
|
415
|
+
|
|
412
416
|
|
|
413
417
|
# ---------------------------------------------------------------------------
|
|
414
418
|
# Ollama — response extraction
|
|
@@ -486,29 +490,27 @@ class TestDumbTransportNoRedaction:
|
|
|
486
490
|
assert "DUMB TRANSPORT" in doc
|
|
487
491
|
assert "do NOT redact" in doc or "does NOT redact" in doc
|
|
488
492
|
|
|
489
|
-
def test_anthropic_passes_sensitive_text_through_unchanged(self
|
|
490
|
-
mock_client
|
|
493
|
+
def test_anthropic_passes_sensitive_text_through_unchanged(self):
|
|
494
|
+
mock_client = _make_anthropic_client()
|
|
491
495
|
anthropic_complete(
|
|
492
|
-
|
|
496
|
+
client=mock_client,
|
|
493
497
|
model="m",
|
|
494
498
|
max_tokens=10,
|
|
495
499
|
system=self.SENSITIVE_TEXT,
|
|
496
500
|
user=self.SENSITIVE_TEXT,
|
|
497
|
-
install_hint="h",
|
|
498
501
|
)
|
|
499
502
|
kwargs = mock_client.messages.create.call_args.kwargs
|
|
500
503
|
assert kwargs["system"] == self.SENSITIVE_TEXT
|
|
501
504
|
assert kwargs["messages"][0]["content"] == self.SENSITIVE_TEXT
|
|
502
505
|
|
|
503
|
-
def test_openai_passes_sensitive_text_through_unchanged(self
|
|
504
|
-
mock_client
|
|
506
|
+
def test_openai_passes_sensitive_text_through_unchanged(self):
|
|
507
|
+
mock_client = _make_openai_client()
|
|
505
508
|
openai_complete(
|
|
506
|
-
|
|
509
|
+
client=mock_client,
|
|
507
510
|
model="m",
|
|
508
511
|
max_tokens=10,
|
|
509
512
|
system=self.SENSITIVE_TEXT,
|
|
510
513
|
user=self.SENSITIVE_TEXT,
|
|
511
|
-
install_hint="h",
|
|
512
514
|
)
|
|
513
515
|
messages = mock_client.chat.completions.create.call_args.kwargs["messages"]
|
|
514
516
|
assert messages[0]["content"] == self.SENSITIVE_TEXT
|
|
@@ -534,6 +536,10 @@ class TestDumbTransportNoRedaction:
|
|
|
534
536
|
|
|
535
537
|
|
|
536
538
|
def test_import_light_no_anthropic_or_openai_loaded():
|
|
539
|
+
"""Stronger than a merely-lazy-import claim: llm.py contains no reference
|
|
540
|
+
to anthropic/openai anywhere (no lazy-import helpers either, since the
|
|
541
|
+
SDK providers now take an injected, duck-typed client) — so the SDKs are
|
|
542
|
+
never loaded into sys.modules just by importing this module."""
|
|
537
543
|
code = (
|
|
538
544
|
"import importlib, sys; "
|
|
539
545
|
"importlib.import_module('shipwright_kit.llm'); "
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
{shipwright_kit-0.10.0 → shipwright_kit-0.11.0}/shipwright_kit.egg-info/dependency_links.txt
RENAMED
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|