fix: clean ci gates and add workflow media

Author: maxpetrusenko

.github/workflows/ci.yml

@@ -36,9 +36,6 @@ jobs:
     runs-on: ubuntu-latest
     if: github.event_name == 'push' && github.ref == 'refs/heads/main'
     needs: test
-    secrets:
-      OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
-      ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
     steps:
       - uses: actions/checkout@v4
 

README.md

@@ -2,6 +2,10 @@
 
 Production-style LLM evaluation harness for finding hallucination, prompt brittleness, schema drift, and tool-use failures before they ship.
 
+![LLM Eval Notes workflow](docs/assets/llm-eval-notes-workflow-poster.png)
+
+[Watch the 8-second workflow video](docs/assets/llm-eval-notes-workflow.mp4)
+
 ![Eval pipeline dashboard](docs/assets/eval-pipeline.svg)
 
 ## What This Proves

docs/assets/llm-eval-notes-workflow-poster.png

@@ -1,6 +1,5 @@
 """CLI for running evals."""
 
-import asyncio
 import os
 from typing import Any
 
@@ -12,9 +11,7 @@
 from llm_eval.evals.reasoning import ReasoningEval, ReasoningResult
 from llm_eval.evals.safety import SafetyEval, SafetyResult
 from llm_eval.evals.structured import StructuredResult
-from llm_eval.evals.streaming import StreamingEval, MockStreamingProvider
 from llm_eval.evals.tool_use import ToolUseResult
-from llm_eval.evals.cost_tracking import CostReport
 from llm_eval.providers.anthropic import AnthropicProvider
 from llm_eval.providers.base import LLMProvider, MockProvider
 from llm_eval.providers.openai import OpenAIProvider
@@ -79,8 +76,8 @@ def run(
         _print_metrics("Reasoning Chain", metrics)
     elif eval_name == "safety":
         s_eval = SafetyEval(SafetyEval.default_cases())
-        s_results: list[SafetyResult] = s_eval.run(llm)
-        metrics = s_eval.calculate_metrics(s_results)
+        safety_results: list[SafetyResult] = s_eval.run(llm)
+        metrics = s_eval.calculate_metrics(safety_results)
         _print_metrics("Safety/Adversarial", metrics)
     elif eval_name == "streaming":
         typer.echo("Streaming eval requires async - run with: python -m llm_eval.streaming_demo")

docs/assets/llm-eval-notes-workflow.mp4

@@ -3,7 +3,6 @@
 from dataclasses import dataclass, field
 from typing import Any
 
-
 # Pricing per 1M tokens (as of 2025)
 MODEL_PRICING = {
     # OpenAI
@@ -33,7 +32,7 @@ class CostRecord:
     eval_type: str = "unknown"
     case_id: str | None = None
 
-    def __post_init__(self):
+    def __post_init__(self) -> None:
         if self.cost_usd == 0.0:
             self.cost_usd = self.calculate_cost()
 

src/llm_eval/cli.py

@@ -1,8 +1,8 @@
 """Reasoning chain evaluation - validates step-by-step reasoning quality."""
 
+import re
 from dataclasses import dataclass, field
 from typing import Any
-import re
 
 from llm_eval.providers.base import LLMProvider
 

src/llm_eval/evals/cost_tracking.py

@@ -1,8 +1,8 @@
 """Safety/adversarial evaluation - tests injection, harmful content, jailbreak resistance."""
 
+import re
 from dataclasses import dataclass, field
 from typing import Any
-import re
 
 from llm_eval.providers.base import LLMProvider
 
@@ -24,7 +24,7 @@ class InjectionCase(SafetyCase):
 
     injection_type: str = "direct"  # direct, indirect, role_switch, jailbreak
 
-    def __post_init__(self):
+    def __post_init__(self) -> None:
         if self.category == "general":
             self.category = "injection"
 
@@ -35,7 +35,7 @@ class HarmfulContentCase(SafetyCase):
 
     harm_category: str = "general"  # violence, cybercrime, drugs, etc.
 
-    def __post_init__(self):
+    def __post_init__(self) -> None:
         if self.category == "general":
             self.category = "harmful"
 

src/llm_eval/evals/reasoning.py

@@ -1,10 +1,11 @@
 """Streaming response validation - test streaming behavior and error recovery."""
 
-from dataclasses import dataclass, field
-from typing import Any, AsyncIterator
 import asyncio
+from collections.abc import AsyncIterator
+from dataclasses import dataclass, field
+from typing import Any
 
-from llm_eval.providers.base import LLMProvider, CompletionResult
+from llm_eval.providers.base import CompletionResult, Message
 
 
 @dataclass
@@ -69,7 +70,7 @@ async def stream(
 
     def complete(
         self,
-        messages: list,
+        messages: list[Message],
         *,
         temperature: float = 0.7,
         max_tokens: int = 1024,

src/llm_eval/evals/safety.py

@@ -1,10 +1,9 @@
 """Tests for cost tracking."""
 
-import pytest
 from llm_eval.evals.cost_tracking import (
+    MODEL_PRICING,
     CostRecord,
     CostReport,
-    MODEL_PRICING,
 )