bridgekit 0.3.9__tar.gz → 0.3.10__tar.gz

{bridgekit-0.3.9 → bridgekit-0.3.10}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: bridgekit
-Version: 0.3.9
+Version: 0.3.10
 Summary: AI tools that make you a better data scientist, not a redundant one.
 License: MIT
 Project-URL: Homepage, https://usebridgekit.com
@@ -398,7 +398,7 @@ willing to commit to — and what's your confidence interval on that estimate?"
 
 ## Tool #5: Compare
 
-Run the same tool through two providers and see both outputs side by side. Useful for evaluating which model works best for your use case — as a one-liner.
+Run the same tool through two providers and see both outputs side by side. A synthesis summary at the top highlights where the models agreed, where they differed in severity, and which gave more actionable feedback — so you get the key insight without reading both outputs in full. Useful for evaluating which model works best for your use case — as a one-liner.
 
 ```python
 from bridgekit import compare
@@ -437,6 +437,16 @@ print(compare(text, providers=["anthropic", "gemini"]))
 BRIDGEKIT COMPARE: EVALUATE
 ─────────────────────────────────────────
 
+SUMMARY
+─────────────────────────────────────────
+Both outputs rated Clarity as STRONG. They diverged on severity: Anthropic
+rated Statistical Rigor as MISSING (harsher) while OpenAI called it NEEDS WORK.
+Anthropic gave more specific feedback — naming the correlation-vs-causation
+problem explicitly and suggesting concrete fixes. OpenAI's feedback stayed
+more generic. Anthropic's bottom line targets the core analytical flaw;
+OpenAI's restates the statistical point only.
+
+
 ANTHROPIC  claude-opus-4-8
 ─────────────────────────────────────────
 BRIDGEKIT ANALYSIS REVIEW
@@ -458,7 +468,61 @@ BRIDGEKIT ANALYSIS REVIEW
 ...
 ```
 
-Both providers are called in parallel, so the total wait time is the slower of the two — not the sum.
+Both providers are called in parallel, so the total wait time is the slower of the two — not the sum. The synthesis summary is a third sequential call made after both outputs are ready.
+
+---
+
+## Tool #6: Summarize
+
+Turn a long analysis writeup, notebook, or report into a short executive summary — the gap between doing the analysis and communicating it to people who won't read the whole thing.
+
+```python
+from bridgekit import summarize
+
+text = """
+I analyzed 90 days of user behavior data to understand what drives subscription
+upgrades. Users who engaged with the reporting feature within their first week
+were 3x more likely to upgrade within 30 days. Sample size was 1,200 users
+across two acquisition channels, with consistent results in both. I recommend
+we prioritize onboarding users to reporting as a growth lever.
+"""
+
+# Default — general business audience
+print(summarize(text))
+
+# Or specify an audience
+print(summarize(text, audience="VP of Marketing"))
+print(summarize(text, audience="board"))
+
+# Override for longer summaries
+print(summarize(text, max_tokens=2048))
+```
+
+**Output:**
+```
+BRIDGEKIT SUMMARY
+─────────────────────────────────────────
+AUDIENCE: VP of Marketing
+
+KEY TAKEAWAY
+Getting users into the reporting feature in their first week is our strongest
+predictor of paid upgrades — users who engage with it are 3x more likely to
+convert within 30 days.
+
+WHAT WE FOUND
+- Early reporting engagement (week 1) drives 3x higher upgrade rates within 30 days
+- Finding holds across both acquisition channels, suggesting it's a genuine
+  behavior pattern, not channel-specific
+- Analyzed 1,200 users over 90 days with sufficient scale to trust the result
+
+SO WHAT
+We should redesign onboarding to get users to reporting faster. This is a
+high-confidence growth lever worth testing immediately.
+
+─────────────────────────────────────────
+```
+
+`audience`, `provider`, `model`, `system_prompt`, and `max_tokens` are all optional — the more specific the audience, the more tailored the summary.
 
 ---
 
@@ -498,6 +562,7 @@ All tools support the same `provider` and `model` parameters:
 - `plan(question, provider=None, model=None, ..., system_prompt=None)`
 - `ask(question, provider=None, model=None, ..., system_prompt=None)`
 - `redteam(text, provider=None, model=None, ..., system_prompt=None)`
+- `summarize(text, provider=None, model=None, ..., system_prompt=None)`
 
 ---
 
@@ -506,7 +571,7 @@ All tools support the same `provider` and `model` parameters:
 Every tool accepts an optional `system_prompt` parameter to override the default persona. Use this to adapt the tone or focus to a specific domain without changing anything else.
 
 ```python
-from bridgekit import evaluate, plan, ask, redteam
+from bridgekit import evaluate, plan, ask, redteam, summarize
 
 # Narrow the reviewer to a specific domain
 print(evaluate("my analysis", system_prompt="You are a skeptical PhD statistician focused only on methodology"))
@@ -519,6 +584,9 @@ print(redteam("my analysis", system_prompt="You are a hostile regulator looking
 
 # Change the answering style for ask
 print(ask("my question", text="...", system_prompt="You are a financial analyst. Answer only in terms of revenue impact."))
+
+# Replace the summarizer persona entirely
+print(summarize("my analysis", system_prompt="You are a data journalist writing a one-paragraph news brief."))
 ```
 
 When `system_prompt` is not provided, each tool uses its built-in default — existing behavior is unchanged.

{bridgekit-0.3.9 → bridgekit-0.3.10}/README.md

@@ -366,7 +366,7 @@ willing to commit to — and what's your confidence interval on that estimate?"
 
 ## Tool #5: Compare
 
-Run the same tool through two providers and see both outputs side by side. Useful for evaluating which model works best for your use case — as a one-liner.
+Run the same tool through two providers and see both outputs side by side. A synthesis summary at the top highlights where the models agreed, where they differed in severity, and which gave more actionable feedback — so you get the key insight without reading both outputs in full. Useful for evaluating which model works best for your use case — as a one-liner.
 
 ```python
 from bridgekit import compare
@@ -405,6 +405,16 @@ print(compare(text, providers=["anthropic", "gemini"]))
 BRIDGEKIT COMPARE: EVALUATE
 ─────────────────────────────────────────
 
+SUMMARY
+─────────────────────────────────────────
+Both outputs rated Clarity as STRONG. They diverged on severity: Anthropic
+rated Statistical Rigor as MISSING (harsher) while OpenAI called it NEEDS WORK.
+Anthropic gave more specific feedback — naming the correlation-vs-causation
+problem explicitly and suggesting concrete fixes. OpenAI's feedback stayed
+more generic. Anthropic's bottom line targets the core analytical flaw;
+OpenAI's restates the statistical point only.
+
+
 ANTHROPIC  claude-opus-4-8
 ─────────────────────────────────────────
 BRIDGEKIT ANALYSIS REVIEW
@@ -426,7 +436,61 @@ BRIDGEKIT ANALYSIS REVIEW
 ...
 ```
 
-Both providers are called in parallel, so the total wait time is the slower of the two — not the sum.
+Both providers are called in parallel, so the total wait time is the slower of the two — not the sum. The synthesis summary is a third sequential call made after both outputs are ready.
+
+---
+
+## Tool #6: Summarize
+
+Turn a long analysis writeup, notebook, or report into a short executive summary — the gap between doing the analysis and communicating it to people who won't read the whole thing.
+
+```python
+from bridgekit import summarize
+
+text = """
+I analyzed 90 days of user behavior data to understand what drives subscription
+upgrades. Users who engaged with the reporting feature within their first week
+were 3x more likely to upgrade within 30 days. Sample size was 1,200 users
+across two acquisition channels, with consistent results in both. I recommend
+we prioritize onboarding users to reporting as a growth lever.
+"""
+
+# Default — general business audience
+print(summarize(text))
+
+# Or specify an audience
+print(summarize(text, audience="VP of Marketing"))
+print(summarize(text, audience="board"))
+
+# Override for longer summaries
+print(summarize(text, max_tokens=2048))
+```
+
+**Output:**
+```
+BRIDGEKIT SUMMARY
+─────────────────────────────────────────
+AUDIENCE: VP of Marketing
+
+KEY TAKEAWAY
+Getting users into the reporting feature in their first week is our strongest
+predictor of paid upgrades — users who engage with it are 3x more likely to
+convert within 30 days.
+
+WHAT WE FOUND
+- Early reporting engagement (week 1) drives 3x higher upgrade rates within 30 days
+- Finding holds across both acquisition channels, suggesting it's a genuine
+  behavior pattern, not channel-specific
+- Analyzed 1,200 users over 90 days with sufficient scale to trust the result
+
+SO WHAT
+We should redesign onboarding to get users to reporting faster. This is a
+high-confidence growth lever worth testing immediately.
+
+─────────────────────────────────────────
+```
+
+`audience`, `provider`, `model`, `system_prompt`, and `max_tokens` are all optional — the more specific the audience, the more tailored the summary.
 
 ---
 
@@ -466,6 +530,7 @@ All tools support the same `provider` and `model` parameters:
 - `plan(question, provider=None, model=None, ..., system_prompt=None)`
 - `ask(question, provider=None, model=None, ..., system_prompt=None)`
 - `redteam(text, provider=None, model=None, ..., system_prompt=None)`
+- `summarize(text, provider=None, model=None, ..., system_prompt=None)`
 
 ---
 
@@ -474,7 +539,7 @@ All tools support the same `provider` and `model` parameters:
 Every tool accepts an optional `system_prompt` parameter to override the default persona. Use this to adapt the tone or focus to a specific domain without changing anything else.
 
 ```python
-from bridgekit import evaluate, plan, ask, redteam
+from bridgekit import evaluate, plan, ask, redteam, summarize
 
 # Narrow the reviewer to a specific domain
 print(evaluate("my analysis", system_prompt="You are a skeptical PhD statistician focused only on methodology"))
@@ -487,6 +552,9 @@ print(redteam("my analysis", system_prompt="You are a hostile regulator looking
 
 # Change the answering style for ask
 print(ask("my question", text="...", system_prompt="You are a financial analyst. Answer only in terms of revenue impact."))
+
+# Replace the summarizer persona entirely
+print(summarize("my analysis", system_prompt="You are a data journalist writing a one-paragraph news brief."))
 ```
 
 When `system_prompt` is not provided, each tool uses its built-in default — existing behavior is unchanged.

{bridgekit-0.3.9 → bridgekit-0.3.10}/bridgekit/__init__.py

@@ -3,6 +3,7 @@ from .search import ask
 from .planner import plan
 from .redteam import redteam
 from .compare import compare
+from .summarize import summarize
 
-__version__ = "0.3.9"
-__all__ = ["evaluate", "ask", "plan", "redteam", "compare"]
+__version__ = "0.3.10"
+__all__ = ["evaluate", "ask", "plan", "redteam", "compare", "summarize"]

bridgekit-0.3.10/bridgekit/summarize.py

@@ -0,0 +1,72 @@
+from .config import DEFAULT_MODEL, parse_provider, get_default_model
+from .providers import create_message
+
+DEFAULT_AUDIENCE = "a general business audience with no technical or data science background"
+
+SYSTEM_PROMPT_TEMPLATE = """You are a senior data scientist preparing an executive summary of a technical analysis for {audience}.
+
+Your job is to translate technical detail into what matters for decision-making. Cut jargon, cut methodology detail unless it's essential to trust the conclusion, and lead with the takeaway. Write for someone who will skim this in 30 seconds.
+
+Format your response exactly like this:
+
+BRIDGEKIT SUMMARY
+─────────────────────────────────────────
+AUDIENCE: {audience_label}
+
+KEY TAKEAWAY
+[1-2 sentences: the single most important finding or recommendation]
+
+WHAT WE FOUND
+[3-5 bullet points of the supporting findings, in plain language]
+
+SO WHAT
+[1-2 sentences on the business implication or recommended next step]
+
+─────────────────────────────────────────
+"""
+
+
+def summarize(text: str, audience: str = None, provider: str = None, model: str = None, system_prompt: str = None, max_tokens: int = 1024) -> str:
+    """
+    Turn a long analysis, notebook, or report into a short executive summary.
+
+    Args:
+        text:          The long-form analysis writeup, notebook output, or report as a plain string.
+        audience:      Optional. Who the summary is for (e.g. "VP of Marketing", "board",
+                       "engineering team"). Defaults to a general business audience.
+        provider:      Optional. The AI provider to use ("anthropic", "openai", "gemini").
+                       If not specified, defaults to "anthropic" or infers from model.
+        model:         Optional. The specific model to use. If not specified, uses the provider's default.
+        system_prompt: Optional. A custom system prompt to fully override the default summarizer persona.
+                       When provided, the audience parameter is ignored.
+        max_tokens:    Optional. Maximum tokens in the response. Defaults to 1024.
+
+    Returns:
+        A short executive summary covering the key takeaway, supporting findings,
+        and the business implication.
+    """
+    if not text or not text.strip():
+        raise ValueError("Text cannot be empty.")
+
+    # Parse provider and determine model
+    provider_enum = parse_provider(provider, model)
+    if model is None:
+        model = get_default_model(provider_enum)
+
+    if system_prompt is None:
+        audience_label = audience if audience else "General Business Audience"
+        audience_desc = audience if audience else DEFAULT_AUDIENCE
+        system_prompt = SYSTEM_PROMPT_TEMPLATE.format(
+            audience=audience_desc,
+            audience_label=audience_label
+        )
+
+    user_message = f"Summarize this analysis:\n\n{text}"
+
+    return create_message(
+        provider=provider_enum,
+        system_prompt=system_prompt,
+        user_message=user_message,
+        model=model,
+        max_tokens=max_tokens
+    )

{bridgekit-0.3.9 → bridgekit-0.3.10}/bridgekit.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: bridgekit
-Version: 0.3.9
+Version: 0.3.10
 Summary: AI tools that make you a better data scientist, not a redundant one.
 License: MIT
 Project-URL: Homepage, https://usebridgekit.com
@@ -398,7 +398,7 @@ willing to commit to — and what's your confidence interval on that estimate?"
 
 ## Tool #5: Compare
 
-Run the same tool through two providers and see both outputs side by side. Useful for evaluating which model works best for your use case — as a one-liner.
+Run the same tool through two providers and see both outputs side by side. A synthesis summary at the top highlights where the models agreed, where they differed in severity, and which gave more actionable feedback — so you get the key insight without reading both outputs in full. Useful for evaluating which model works best for your use case — as a one-liner.
 
 ```python
 from bridgekit import compare
@@ -437,6 +437,16 @@ print(compare(text, providers=["anthropic", "gemini"]))
 BRIDGEKIT COMPARE: EVALUATE
 ─────────────────────────────────────────
 
+SUMMARY
+─────────────────────────────────────────
+Both outputs rated Clarity as STRONG. They diverged on severity: Anthropic
+rated Statistical Rigor as MISSING (harsher) while OpenAI called it NEEDS WORK.
+Anthropic gave more specific feedback — naming the correlation-vs-causation
+problem explicitly and suggesting concrete fixes. OpenAI's feedback stayed
+more generic. Anthropic's bottom line targets the core analytical flaw;
+OpenAI's restates the statistical point only.
+
+
 ANTHROPIC  claude-opus-4-8
 ─────────────────────────────────────────
 BRIDGEKIT ANALYSIS REVIEW
@@ -458,7 +468,61 @@ BRIDGEKIT ANALYSIS REVIEW
 ...
 ```
 
-Both providers are called in parallel, so the total wait time is the slower of the two — not the sum.
+Both providers are called in parallel, so the total wait time is the slower of the two — not the sum. The synthesis summary is a third sequential call made after both outputs are ready.
+
+---
+
+## Tool #6: Summarize
+
+Turn a long analysis writeup, notebook, or report into a short executive summary — the gap between doing the analysis and communicating it to people who won't read the whole thing.
+
+```python
+from bridgekit import summarize
+
+text = """
+I analyzed 90 days of user behavior data to understand what drives subscription
+upgrades. Users who engaged with the reporting feature within their first week
+were 3x more likely to upgrade within 30 days. Sample size was 1,200 users
+across two acquisition channels, with consistent results in both. I recommend
+we prioritize onboarding users to reporting as a growth lever.
+"""
+
+# Default — general business audience
+print(summarize(text))
+
+# Or specify an audience
+print(summarize(text, audience="VP of Marketing"))
+print(summarize(text, audience="board"))
+
+# Override for longer summaries
+print(summarize(text, max_tokens=2048))
+```
+
+**Output:**
+```
+BRIDGEKIT SUMMARY
+─────────────────────────────────────────
+AUDIENCE: VP of Marketing
+
+KEY TAKEAWAY
+Getting users into the reporting feature in their first week is our strongest
+predictor of paid upgrades — users who engage with it are 3x more likely to
+convert within 30 days.
+
+WHAT WE FOUND
+- Early reporting engagement (week 1) drives 3x higher upgrade rates within 30 days
+- Finding holds across both acquisition channels, suggesting it's a genuine
+  behavior pattern, not channel-specific
+- Analyzed 1,200 users over 90 days with sufficient scale to trust the result
+
+SO WHAT
+We should redesign onboarding to get users to reporting faster. This is a
+high-confidence growth lever worth testing immediately.
+
+─────────────────────────────────────────
+```
+
+`audience`, `provider`, `model`, `system_prompt`, and `max_tokens` are all optional — the more specific the audience, the more tailored the summary.
 
 ---
 
@@ -498,6 +562,7 @@ All tools support the same `provider` and `model` parameters:
 - `plan(question, provider=None, model=None, ..., system_prompt=None)`
 - `ask(question, provider=None, model=None, ..., system_prompt=None)`
 - `redteam(text, provider=None, model=None, ..., system_prompt=None)`
+- `summarize(text, provider=None, model=None, ..., system_prompt=None)`
 
 ---
 
@@ -506,7 +571,7 @@ All tools support the same `provider` and `model` parameters:
 Every tool accepts an optional `system_prompt` parameter to override the default persona. Use this to adapt the tone or focus to a specific domain without changing anything else.
 
 ```python
-from bridgekit import evaluate, plan, ask, redteam
+from bridgekit import evaluate, plan, ask, redteam, summarize
 
 # Narrow the reviewer to a specific domain
 print(evaluate("my analysis", system_prompt="You are a skeptical PhD statistician focused only on methodology"))
@@ -519,6 +584,9 @@ print(redteam("my analysis", system_prompt="You are a hostile regulator looking
 
 # Change the answering style for ask
 print(ask("my question", text="...", system_prompt="You are a financial analyst. Answer only in terms of revenue impact."))
+
+# Replace the summarizer persona entirely
+print(summarize("my analysis", system_prompt="You are a data journalist writing a one-paragraph news brief."))
 ```
 
 When `system_prompt` is not provided, each tool uses its built-in default — existing behavior is unchanged.

{bridgekit-0.3.9 → bridgekit-0.3.10}/bridgekit.egg-info/SOURCES.txt

@@ -10,6 +10,7 @@ bridgekit/providers.py
 bridgekit/redteam.py
 bridgekit/reviewer.py
 bridgekit/search.py
+bridgekit/summarize.py
 bridgekit.egg-info/PKG-INFO
 bridgekit.egg-info/SOURCES.txt
 bridgekit.egg-info/dependency_links.txt
@@ -23,4 +24,5 @@ tests/test_planner.py
 tests/test_providers.py
 tests/test_redteam.py
 tests/test_reviewer.py
-tests/test_search.py
+tests/test_search.py
+tests/test_summarize.py

{bridgekit-0.3.9 → bridgekit-0.3.10}/pyproject.toml

@@ -7,7 +7,7 @@ include = ["bridgekit*"]
 
 [project]
 name = "bridgekit"
-version = "0.3.9"
+version = "0.3.10"
 description = "AI tools that make you a better data scientist, not a redundant one."
 readme = "README.md"
 requires-python = ">=3.9"

bridgekit-0.3.10/tests/test_summarize.py

@@ -0,0 +1,212 @@
+import os
+import pytest
+from unittest.mock import MagicMock, patch
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+def _make_mock_message(text: str):
+    content_block = MagicMock()
+    content_block.text = text
+    message = MagicMock()
+    message.content = [content_block]
+    return message
+
+
+FAKE_RESPONSE = (
+    "BRIDGEKIT SUMMARY\n"
+    "─────────────────────────────────────────\n"
+    "AUDIENCE: General Business Audience\n\n"
+    "KEY TAKEAWAY\n"
+    "The new onboarding flow increased upgrade rates by 3x within 30 days.\n\n"
+    "WHAT WE FOUND\n"
+    "- Users who engaged with reporting in week 1 were 3x more likely to upgrade\n"
+    "- The effect held across acquisition channels\n"
+    "- Sample size was 500 users\n\n"
+    "SO WHAT\n"
+    "Prioritize onboarding users to the reporting feature as a growth lever.\n"
+    "─────────────────────────────────────────\n"
+)
+
+
+# ---------------------------------------------------------------------------
+# Tests
+# ---------------------------------------------------------------------------
+
+class TestSummarizeReturnsString:
+    """summarize() should return a non-empty string."""
+
+    def test_returns_string(self):
+        with patch.dict(os.environ, {"ANTHROPIC_API_KEY": "test-key"}):
+            with patch("anthropic.Anthropic") as MockAnthropic:
+                mock_client = MagicMock()
+                mock_client.messages.create.return_value = _make_mock_message(FAKE_RESPONSE)
+                MockAnthropic.return_value = mock_client
+
+                from bridgekit.summarize import summarize
+                result = summarize("We ran an A/B test on 500 users and saw a 3x lift in upgrades.")
+
+        assert isinstance(result, str)
+        assert len(result) > 0
+
+
+class TestSummarizeOutputStructure:
+    """summarize() output should contain the required section headers."""
+
+    def test_output_contains_key_takeaway(self):
+        with patch.dict(os.environ, {"ANTHROPIC_API_KEY": "test-key"}):
+            with patch("anthropic.Anthropic") as MockAnthropic:
+                mock_client = MagicMock()
+                mock_client.messages.create.return_value = _make_mock_message(FAKE_RESPONSE)
+                MockAnthropic.return_value = mock_client
+
+                from bridgekit.summarize import summarize
+                result = summarize("Some analysis text.")
+
+        assert "KEY TAKEAWAY" in result
+
+    def test_output_contains_so_what(self):
+        with patch.dict(os.environ, {"ANTHROPIC_API_KEY": "test-key"}):
+            with patch("anthropic.Anthropic") as MockAnthropic:
+                mock_client = MagicMock()
+                mock_client.messages.create.return_value = _make_mock_message(FAKE_RESPONSE)
+                MockAnthropic.return_value = mock_client
+
+                from bridgekit.summarize import summarize
+                result = summarize("Some analysis text.")
+
+        assert "SO WHAT" in result
+
+
+class TestSummarizeMissingApiKey:
+    """summarize() should raise EnvironmentError when the API key is absent."""
+
+    def test_raises_environment_error_when_key_missing(self):
+        env = {k: v for k, v in os.environ.items() if k != "ANTHROPIC_API_KEY"}
+        with patch.dict(os.environ, env, clear=True):
+            from bridgekit.summarize import summarize
+            with pytest.raises(EnvironmentError):
+                summarize("Some analysis text.")
+
+    def test_error_message_mentions_key(self):
+        env = {k: v for k, v in os.environ.items() if k != "ANTHROPIC_API_KEY"}
+        with patch.dict(os.environ, env, clear=True):
+            from bridgekit.summarize import summarize
+            with pytest.raises(EnvironmentError, match="ANTHROPIC_API_KEY"):
+                summarize("Some analysis text.")
+
+
+class TestSummarizeEmptyInput:
+    """summarize() should raise ValueError for empty or whitespace-only input."""
+
+    def test_empty_string_raises_value_error(self):
+        with patch.dict(os.environ, {"ANTHROPIC_API_KEY": "test-key"}):
+            from bridgekit.summarize import summarize
+            with pytest.raises(ValueError, match="empty"):
+                summarize("")
+
+    def test_whitespace_only_raises_value_error(self):
+        with patch.dict(os.environ, {"ANTHROPIC_API_KEY": "test-key"}):
+            from bridgekit.summarize import summarize
+            with pytest.raises(ValueError, match="empty"):
+                summarize("   ")
+
+
+class TestSummarizeAudience:
+    """summarize() should include a custom audience in the system prompt."""
+
+    def test_custom_audience_reaches_system_prompt(self):
+        with patch.dict(os.environ, {"ANTHROPIC_API_KEY": "test-key"}):
+            with patch("anthropic.Anthropic") as MockAnthropic:
+                mock_client = MagicMock()
+                mock_client.messages.create.return_value = _make_mock_message(FAKE_RESPONSE)
+                MockAnthropic.return_value = mock_client
+
+                from bridgekit.summarize import summarize
+                summarize("Some analysis text.", audience="VP of Marketing")
+
+                call_kwargs = mock_client.messages.create.call_args
+                assert "VP of Marketing" in call_kwargs.kwargs.get("system", "")
+
+    def test_default_audience_used_when_not_specified(self):
+        with patch.dict(os.environ, {"ANTHROPIC_API_KEY": "test-key"}):
+            with patch("anthropic.Anthropic") as MockAnthropic:
+                mock_client = MagicMock()
+                mock_client.messages.create.return_value = _make_mock_message(FAKE_RESPONSE)
+                MockAnthropic.return_value = mock_client
+
+                from bridgekit.summarize import summarize
+                summarize("Some analysis text.")
+
+                call_kwargs = mock_client.messages.create.call_args
+                assert "General Business Audience" in call_kwargs.kwargs.get("system", "")
+
+
+class TestSummarizeCustomSystemPrompt:
+    """summarize() should forward a custom system_prompt to the API, ignoring audience."""
+
+    def test_custom_system_prompt_reaches_api(self):
+        custom_prompt = "You are a data journalist writing a one-paragraph news brief."
+        with patch.dict(os.environ, {"ANTHROPIC_API_KEY": "test-key"}):
+            with patch("anthropic.Anthropic") as MockAnthropic:
+                mock_client = MagicMock()
+                mock_client.messages.create.return_value = _make_mock_message(FAKE_RESPONSE)
+                MockAnthropic.return_value = mock_client
+
+                from bridgekit.summarize import summarize
+                summarize("Some analysis text.", system_prompt=custom_prompt)
+
+                call_kwargs = mock_client.messages.create.call_args
+                assert call_kwargs.kwargs.get("system") == custom_prompt
+
+
+class TestSummarizeApiCallShape:
+    """summarize() should pass the user text through to the Anthropic API."""
+
+    def test_api_called_with_user_text(self):
+        user_text = "Our conversion rate improved after the campaign."
+        with patch.dict(os.environ, {"ANTHROPIC_API_KEY": "test-key"}):
+            with patch("anthropic.Anthropic") as MockAnthropic:
+                mock_client = MagicMock()
+                mock_client.messages.create.return_value = _make_mock_message(FAKE_RESPONSE)
+                MockAnthropic.return_value = mock_client
+
+                from bridgekit.summarize import summarize
+                summarize(user_text)
+
+                call_kwargs = mock_client.messages.create.call_args
+                messages_arg = call_kwargs.kwargs.get("messages") or call_kwargs.args[0]
+                content = str(messages_arg)
+                assert user_text in content
+
+
+class TestSummarizeMaxTokens:
+    """summarize() should pass max_tokens through to the API."""
+
+    def test_default_max_tokens_is_1024(self):
+        with patch.dict(os.environ, {"ANTHROPIC_API_KEY": "test-key"}):
+            with patch("anthropic.Anthropic") as MockAnthropic:
+                mock_client = MagicMock()
+                mock_client.messages.create.return_value = _make_mock_message(FAKE_RESPONSE)
+                MockAnthropic.return_value = mock_client
+
+                from bridgekit.summarize import summarize
+                summarize("Some analysis text.")
+
+                call_kwargs = mock_client.messages.create.call_args
+                assert call_kwargs.kwargs.get("max_tokens") == 1024
+
+    def test_custom_max_tokens_reaches_api(self):
+        with patch.dict(os.environ, {"ANTHROPIC_API_KEY": "test-key"}):
+            with patch("anthropic.Anthropic") as MockAnthropic:
+                mock_client = MagicMock()
+                mock_client.messages.create.return_value = _make_mock_message(FAKE_RESPONSE)
+                MockAnthropic.return_value = mock_client
+
+                from bridgekit.summarize import summarize
+                summarize("Some analysis text.", max_tokens=2048)
+
+                call_kwargs = mock_client.messages.create.call_args
+                assert call_kwargs.kwargs.get("max_tokens") == 2048