agent-eval-flow 0.5.1__tar.gz

agent_eval_flow-0.5.1/.gitattributes

@@ -0,0 +1,6 @@
+# The acceptance checkpoint hashes exact original bytes, including upstream logs.
+tests/** -text
+pytest.ini -text
+requirements-test.txt -text
+# Workflow derivation receipts pin exact bytes; never normalize their line endings.
+examples/workflows/*.json -text whitespace=cr-at-eol

agent_eval_flow-0.5.1/.github/ISSUE_TEMPLATE/bug_report.yml

@@ -0,0 +1,65 @@
+{
+  "name": "Bug report",
+  "description": "Report unexpected behavior with a minimal reproduction.",
+  "body": [
+    {
+      "type": "markdown",
+      "attributes": {
+        "value": "Small, reproducible examples are most useful. Remove credentials, private task data, and sensitive prompts before sharing anything. Raw run logs are not required."
+      }
+    },
+    {
+      "type": "textarea",
+      "id": "environment",
+      "attributes": {
+        "label": "Versions and runtime",
+        "description": "Include the package version or commit, Python version, OS, and relevant runtime or importer. Use not applicable where appropriate.",
+        "placeholder": "Agent Eval Flow:\nPython / OS:\nRuntime, adapter or import format (and revision, if known):"
+      },
+      "validations": {
+        "required": true
+      }
+    },
+    {
+      "type": "textarea",
+      "id": "reproduction",
+      "attributes": {
+        "label": "Minimal reproduction",
+        "description": "Give the smallest code, configuration and steps that reproduce the problem. Use public or redacted sample data."
+      },
+      "validations": {
+        "required": true
+      }
+    },
+    {
+      "type": "textarea",
+      "id": "expected",
+      "attributes": {
+        "label": "Expected behavior",
+        "description": "What result or behavior did you expect?"
+      },
+      "validations": {
+        "required": true
+      }
+    },
+    {
+      "type": "textarea",
+      "id": "observed",
+      "attributes": {
+        "label": "Observed behavior",
+        "description": "What happened instead? Include relevant statuses or error messages."
+      },
+      "validations": {
+        "required": true
+      }
+    },
+    {
+      "type": "textarea",
+      "id": "evidence",
+      "attributes": {
+        "label": "Scrubbed evidence (optional)",
+        "description": "A short traceback, report excerpt or screenshot can help. Prefer the relevant excerpt over a complete capture, and remove sensitive content."
+      }
+    }
+  ]
+}

agent_eval_flow-0.5.1/.github/ISSUE_TEMPLATE/feature_request.yml

@@ -0,0 +1,58 @@
+{
+  "name": "Feature or use case",
+  "description": "Explain a workflow, missing capability, or useful improvement.",
+  "body": [
+    {
+      "type": "markdown",
+      "attributes": {
+        "value": "Describe the decision or workflow you want to support. A specific example helps, but private data and raw run logs are not needed."
+      }
+    },
+    {
+      "type": "textarea",
+      "id": "use_case",
+      "attributes": {
+        "label": "Problem or use case",
+        "description": "What are you evaluating, and what decision are you trying to make?"
+      },
+      "validations": {
+        "required": true
+      }
+    },
+    {
+      "type": "textarea",
+      "id": "desired_outcome",
+      "attributes": {
+        "label": "Desired outcome",
+        "description": "What would a useful result or supported workflow look like? An example input and output can help."
+      },
+      "validations": {
+        "required": true
+      }
+    },
+    {
+      "type": "textarea",
+      "id": "current_workflow",
+      "attributes": {
+        "label": "Current workflow (optional)",
+        "description": "Mention the runtime or imported format, relevant versions, and your current approach or workaround."
+      }
+    },
+    {
+      "type": "textarea",
+      "id": "proposal",
+      "attributes": {
+        "label": "Proposed approach or alternatives (optional)",
+        "description": "Share a possible solution, alternatives, or constraints. A proposed implementation is not required."
+      }
+    },
+    {
+      "type": "textarea",
+      "id": "examples",
+      "attributes": {
+        "label": "Public or redacted examples (optional)",
+        "description": "Link related issues or include a small example. Remove credentials, private prompts and task data."
+      }
+    }
+  ]
+}

agent_eval_flow-0.5.1/.github/pull_request_template.md

@@ -0,0 +1,15 @@
+## What changed
+
+<!-- Describe the problem and resulting behavior. Link related issues if relevant. -->
+
+## Validation
+
+<!-- List checks and their outcomes, including anything not run and why.
+For a bug fix, include the regression check that demonstrates the behavior.
+See CONTRIBUTING.md for the local checks. -->
+
+## Notes for review
+
+<!-- Optional: explain changes to metric meaning, evidence/provenance,
+missing values, or compatibility. Note remaining limitations.
+Keep credentials, private task data and raw local captures out of the PR. -->

agent_eval_flow-0.5.1/.github/workflows/publish.yml

@@ -0,0 +1,136 @@
+name: Publish to PyPI
+
+on:
+  release:
+    types: [published]
+  # Retry a published release with: gh workflow run publish.yml --ref v0.5.1
+  # Dispatching a branch never enables either job.
+  workflow_dispatch:
+
+permissions: {}
+
+concurrency:
+  group: pypi-${{ github.ref }}
+  cancel-in-progress: false
+
+jobs:
+  build:
+    if: github.repository == 'guybass/agent-eval-flow' && startsWith(github.ref, 'refs/tags/v')
+    runs-on: ubuntu-latest
+    timeout-minutes: 15
+    permissions:
+      contents: read
+      actions: read
+    outputs:
+      artifact-id: ${{ steps.distributions.outputs.artifact-id }}
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
+        with:
+          ref: ${{ github.sha }}
+          persist-credentials: false
+      - uses: actions/setup-python@5fda3b95a4ea91299a34e894583c3862153e4b97 # v7
+        with:
+          python-version: '3.13'
+      - name: Verify published release, source version, and all six CI jobs
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          python - <<'PY'
+          import json
+          import os
+          from pathlib import Path
+          import re
+          import subprocess
+          import tomllib
+          from urllib.parse import urlencode
+          from urllib.request import Request, urlopen
+
+          def require(condition, message):
+              if not condition:
+                  raise SystemExit(message)
+
+          repository = os.environ['GITHUB_REPOSITORY']
+          tag = os.environ['GITHUB_REF_NAME']
+          sha = os.environ['GITHUB_SHA']
+          require(repository == 'guybass/agent-eval-flow', 'Unexpected repository')
+          require(os.environ['GITHUB_REF'] == f'refs/tags/{tag}', 'A version tag is required')
+          require(re.fullmatch(r'v(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)', tag),
+                  'Only final vMAJOR.MINOR.PATCH release tags are publishable')
+          require(os.environ['GITHUB_EVENT_NAME'] in {'release', 'workflow_dispatch'},
+                  'Unexpected publishing event')
+
+          def api(path):
+              request = Request(f'https://api.github.com/repos/{repository}/{path}', headers={
+                  'Authorization': f'Bearer {os.environ["GH_TOKEN"]}',
+                  'Accept': 'application/vnd.github+json',
+                  'X-GitHub-Api-Version': '2022-11-28',
+              })
+              with urlopen(request, timeout=30) as response:
+                  return json.load(response)
+
+          release = api(f'releases/tags/{tag}')
+          require(release['tag_name'] == tag and not release['draft']
+                  and not release['prerelease'] and release['published_at'],
+                  'The tag must have a published, non-prerelease GitHub release')
+          require(api(f'commits/{tag}')['sha'] == sha, 'Tag moved or event SHA differs')
+          require(subprocess.check_output(['git', 'rev-parse', 'HEAD'], text=True).strip() == sha,
+                  'Checkout differs from the verified release commit')
+          require(api(f'compare/{sha}...main')['status'] in {'ahead', 'identical'},
+                  'Release commit must belong to main history')
+          project = tomllib.loads(Path('pyproject.toml').read_text(encoding='utf-8'))['project']
+          require(project['name'] == 'agent-eval-flow' and project['version'] == tag[1:],
+                  'Project name/version must match the release tag exactly')
+
+          query = urlencode({'head_sha': sha, 'event': 'push', 'branch': 'main', 'per_page': 100})
+          runs = api(f'actions/workflows/tests.yml/runs?{query}')['workflow_runs']
+          require(runs, 'No Tests and package run exists for this commit on main')
+          run = max(runs, key=lambda row: row['id'])
+          require(run['head_sha'] == sha and run['head_branch'] == 'main'
+                  and run['event'] == 'push' and run['status'] == 'completed'
+                  and run['conclusion'] == 'success',
+                  'Latest Tests and package run for this commit must finish successfully')
+          jobs = api(f'actions/runs/{run["id"]}/jobs?filter=latest&per_page=100')
+          expected = {f'offline ({system}, {version})'
+                      for system in ('ubuntu-latest', 'windows-latest')
+                      for version in ('3.11', '3.12', '3.13')}
+          require(jobs['total_count'] == len(expected) and len(jobs['jobs']) == len(expected)
+                  and {job['name'] for job in jobs['jobs']} == expected
+                  and all(job['status'] == 'completed' and job['conclusion'] == 'success'
+                          for job in jobs['jobs']),
+                  'All six expected Python/OS matrix jobs must pass; skipped jobs do not qualify')
+          print(f'Verified {tag} at {sha}; CI: {run["html_url"]}')
+          PY
+      - name: Build and validate distributions
+        run: |
+          python -m pip install build twine
+          python -m build
+          python -m twine check --strict dist/*
+          python scripts/verify_distribution.py
+      - uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        id: distributions
+        with:
+          name: distributions-${{ github.ref_name }}
+          path: dist/*
+          if-no-files-found: error
+          retention-days: 7
+
+  publish:
+    needs: build
+    if: github.repository == 'guybass/agent-eval-flow' && startsWith(github.ref, 'refs/tags/v')
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    environment:
+      name: pypi
+      url: https://pypi.org/project/agent-eval-flow/
+    permissions:
+      id-token: write
+    steps:
+      - uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
+        with:
+          artifact-ids: ${{ needs.build.outputs.artifact-id }}
+          path: dist/
+          digest-mismatch: error
+      - name: Publish verified distributions with Trusted Publishing
+        uses: pypa/gh-action-pypi-publish@dc37677b2e1c63e2034f94d8a5b11f265b73ba33 # v1.14.2
+        with:
+          attestations: true

agent_eval_flow-0.5.1/.github/workflows/tests.yml

@@ -0,0 +1,32 @@
+name: Tests and package
+
+on:
+  push:
+  pull_request:
+
+permissions:
+  contents: read
+
+jobs:
+  offline:
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, windows-latest]
+        python: ['3.11', '3.12', '3.13']
+    runs-on: ${{ matrix.os }}
+    timeout-minutes: 15
+    steps:
+      - uses: actions/checkout@v7
+        with:
+          persist-credentials: false
+      - uses: actions/setup-python@v7
+        with:
+          python-version: ${{ matrix.python }}
+          cache: pip
+      - run: python -m pip install -e ".[test,cli]" build
+      - run: python scripts/verify_acceptance_checkpoint.py
+      - run: python -m pytest --junitxml=test-artifacts/pytest.xml
+      - run: python examples/archive_review.py --output demo-output
+      - run: python -m build
+      - run: python scripts/verify_distribution.py

agent_eval_flow-0.5.1/CONTRIBUTING.md

@@ -0,0 +1,30 @@
+# Contributing
+
+Agent Eval Flow is a developer preview. Issues and small pull requests are welcome.
+For a bug, include the package/Python versions, a minimal reproducible example,
+the expected outcome and the observed result. Remove credentials and private
+task data before sharing evidence.
+
+## Local checks
+
+From a checkout, using Python 3.11 or later:
+
+```bash
+python -m pip install -e ".[test,cli]" build
+python scripts/verify_acceptance_checkpoint.py
+python -m pytest
+python examples/archive_review.py --output demo-output
+python -m build
+python scripts/verify_distribution.py
+```
+
+Default tests do not start live models. Live profiles require explicit selection
+and prepared runtimes; their setup is documented in [tests/e2e/SHOWCASE.md](tests/e2e/SHOWCASE.md).
+
+Keep metric meaning, provenance and missingness explicit. Fixes should include
+a regression check that demonstrates the affected behavior. Preserve the frozen
+acceptance fixtures and their hashes; add new fixtures separately.
+
+Local outputs belong in ignored `demo-output/` or `test-artifacts/` directories.
+Include only deliberately selected, reviewed report artifacts in documentation.
+Third-party fixtures retain their own licenses and provenance.

agent_eval_flow-0.5.1/MANIFEST.in

@@ -0,0 +1,8 @@
+include README.md CONTRIBUTING.md THIRD_PARTY_NOTICES.md pyproject.toml pytest.ini requirements-test.txt .gitattributes
+graft .github
+graft docs
+graft tests
+graft examples
+graft scripts
+global-exclude *.pyc *.pyo
+global-exclude __pycache__

agent_eval_flow-0.5.1/PKG-INFO

@@ -0,0 +1,197 @@
+Metadata-Version: 2.4
+Name: agent-eval-flow
+Version: 0.5.1
+Summary: Evidence-preserving evaluation of complete agent systems
+Project-URL: Homepage, https://guybass.github.io/agent-eval-flow/
+Project-URL: Repository, https://github.com/guybass/agent-eval-flow
+Project-URL: Issues, https://github.com/guybass/agent-eval-flow/issues
+Project-URL: Documentation, https://github.com/guybass/agent-eval-flow/blob/main/README.md
+Project-URL: Examples, https://guybass.github.io/agent-eval-flow/
+Project-URL: Changelog, https://github.com/guybass/agent-eval-flow/releases
+Keywords: agents,agent-evaluation,llm,evaluation,execution-evidence
+Classifier: Development Status :: 3 - Alpha
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Requires-Python: >=3.11
+Description-Content-Type: text/markdown
+Requires-Dist: pydantic<3,>=2.13.5
+Requires-Dist: anyio<5,>=4.15.1
+Requires-Dist: jinja2<4,>=3.1.6
+Requires-Dist: typing-extensions>=4.14
+Provides-Extra: test
+Requires-Dist: pytest<10,>=8; extra == "test"
+Requires-Dist: jsonschema<5,>=4.26; extra == "test"
+Provides-Extra: cli
+Requires-Dist: jsonschema<5,>=4.26; extra == "cli"
+
+[![Agent Eval Flow — capture evidence, apply checks, compare a change](https://raw.githubusercontent.com/guybass/agent-eval-flow/main/docs/assets/agent-eval-flow-banner.png)](https://guybass.github.io/agent-eval-flow/)
+
+# Agent Eval Flow
+
+[![Release](https://img.shields.io/github/v/release/guybass/agent-eval-flow?include_prereleases&label=release&color=245c50)](https://github.com/guybass/agent-eval-flow/releases)
+[![Tests and package](https://github.com/guybass/agent-eval-flow/actions/workflows/tests.yml/badge.svg?branch=main)](https://github.com/guybass/agent-eval-flow/actions/workflows/tests.yml)
+[![Python 3.11+](https://img.shields.io/badge/python-3.11%2B-49659c?logo=python&logoColor=white)](https://github.com/guybass/agent-eval-flow/blob/main/pyproject.toml)
+[![Example reports](https://img.shields.io/badge/reports-OpenSRE%20%2B%20OpenKritt-786495)](https://guybass.github.io/agent-eval-flow/)
+
+**Turn agent runs into evidence you can use to improve the system.**
+
+Evaluate complete agent setups: models, instructions, skills, tools, loops,
+memory and environment. Run through an existing runtime or import saved logs,
+apply your checks, and compare changes in a human-readable report.
+
+**[Browse the reports](https://guybass.github.io/agent-eval-flow/)** ·
+**[Try the offline example](#try-it)** ·
+**[Read the design](https://github.com/guybass/agent-eval-flow/blob/main/docs/lld/README.md)** ·
+**[Report an issue](https://github.com/guybass/agent-eval-flow/issues)**
+
+This is a developer preview. Missing evidence stays unknown; configuration
+inspection and behavioral evaluation can run in parallel. The Python package
+is `agent_eval_flow`.
+
+## Try it
+
+```bash
+git clone https://github.com/guybass/agent-eval-flow.git
+cd agent-eval-flow
+python -m pip install -e ".[test,cli]"
+python -m pytest
+python examples/archive_review.py --output demo-output
+```
+
+The example imports a **real, downloaded SWE-agent repair history** containing
+12 actions, failed edits, corrections and a submitted patch. It produces:
+
+- `demo-output/report.html`: task results, diagnostics and linked native evidence;
+- `demo-output/result/`: the typed saved evaluation;
+- `demo-output/regraded/`: a changed evaluation over the same saved execution.
+
+It makes no model calls and never executes commands from the archive. The
+[example source](https://github.com/guybass/agent-eval-flow/blob/main/examples/archive_review.py) shows an application-owned importer
+and custom metrics using the production library.
+
+Requires Python 3.11 or later. The default tests and archive example need no
+model credentials. Native agent runs require the integration-specific setup below.
+
+## Example reports
+
+Two real agent workflows, controlled synthetic tasks, and specific changes
+measured from retained runs. Click a report to open it in your browser.
+
+| OpenSRE: recovery and stopping | OpenKritt: executed proof |
+| --- | --- |
+| [![OpenSRE before-and-after evaluation report](https://raw.githubusercontent.com/guybass/agent-eval-flow/main/docs/examples/agent-evaluation/01-opensre.png)](https://guybass.github.io/agent-eval-flow/examples/agent-evaluation/01-opensre.html) | [![OpenKritt before-and-after evaluation report](https://raw.githubusercontent.com/guybass/agent-eval-flow/main/docs/examples/agent-evaluation/02-openkritt.png)](https://guybass.github.io/agent-eval-flow/examples/agent-evaluation/02-openkritt.html) |
+| Safe recovery stayed **2/2**; post-report calls fell **1 → 0** after a terminal-tool binding fix. | The demo grounding contract passed **0/1 → 1/1** after requiring execution and clarifying the reporting contract. |
+| [Read the report](https://guybass.github.io/agent-eval-flow/examples/agent-evaluation/01-opensre.html) · [Task and fix walkthrough](https://guybass.github.io/agent-eval-flow/examples/agent-evaluation/01-opensre-walkthrough.html) | [Read the report](https://guybass.github.io/agent-eval-flow/examples/agent-evaluation/02-openkritt.html) · [Task and fix walkthrough](https://guybass.github.io/agent-eval-flow/examples/agent-evaluation/02-openkritt-walkthrough.html) |
+
+These are small development experiments with synthetic tasks, not general
+reliability benchmarks. The gallery contains presentation reports; raw local
+run captures and unpublished social drafts are excluded. The offline example
+above is the reproducible starting point for trying the evaluation API.
+See the [case details, run identifiers and limitations](https://github.com/guybass/agent-eval-flow/blob/main/docs/examples/agent-evaluation/README.md).
+
+## Configure once, then evaluate
+
+```python
+from agent_eval_flow import EvaluationPipeline, EvaluationResult
+
+pipeline = EvaluationPipeline(
+    study=study,                 # data, candidates, execution policy and suite
+    backends=backends,            # existing agent runtimes
+    evaluators=evaluators,        # your metric implementations
+)
+result = pipeline.eval()
+result.save("results/experiment")
+result.report("results/experiment.html")
+
+saved = EvaluationResult.load("results/experiment")
+explanation = saved.explain(saved.runs.runs[0].id)
+comparison = saved.compare("baseline", "challenger", metrics=("success_rate",))
+selection = saved.select(policy)  # change cost/latency/quality priorities
+
+# Regrade captured runs without binding or invoking an agent backend.
+regraded = EvaluationPipeline(study=changed_study, evaluators=evaluators).eval(runs=saved.runs)
+```
+
+In an async application, use `await pipeline.aeval()`. Construction and planning
+do not start an agent. Changing the selection policy does not rerun agents or
+metrics. Comparisons currently provide descriptive differences; they do not
+invent confidence intervals.
+
+## The six objects
+
+| Object | Responsibility |
+| --- | --- |
+| `Study` | Question, candidates, execution conditions and evaluation suite |
+| `EvalDataset` | Keyed task tables, public inputs and private evaluator references |
+| `Candidate` | Model, skill, tool, flow and harness configuration |
+| `RunSet` | Every assignment, output, execution, event, resource observation and failure |
+| `EvalSuite` | Versioned metrics, dependencies, score rules and summaries |
+| `EvaluationResult` | Saved measurements, explanations, comparisons and selection |
+
+Pydantic validates the shared records. AnyIO coordinates runtime calls. Jinja2
+renders self-contained reports. Native agent frameworks keep their loops and
+schedulers. Our code owns the common evidence and comparison contracts.
+
+Missing usage remains unknown, failed work remains in the assignment inventory,
+and shared grading activities are counted once. A captured JSON `null` is distinct
+from absent output. Detail rows explain tasks; they do not inflate the sample size.
+
+## Runtime evidence checks
+
+Use the same optional observation contract to check instructions, tools, model
+selection, loops, memory and environment state. Versioned collectors retain
+declared and observed values with phase, boundary, coverage and evidence.
+Reusable checks run through the existing evaluation pipeline, including saved
+run regrading; reports show the expected and observed values. Missing evidence
+stays unknown. See the [runtime evidence guide](https://github.com/guybass/agent-eval-flow/blob/main/docs/implementation/runtime_evidence.md).
+
+## Runtime integrations
+
+Concrete adapter modules cover Codex, Claude Code, OpenSRE, OpenKritt, Harbor,
+SkillEvaluator imports and NeMo Agent Toolkit batch grading. Their native
+capture/lifecycle code is separate from the core. Prepared service connections,
+version-specific configuration and deployment credentials remain runtime bindings.
+See [implementation status and setup](https://github.com/guybass/agent-eval-flow/blob/main/docs/IMPLEMENTATION.md) for the supported
+boundaries and the live checks still required.
+
+OpenSRE and OpenKritt are separate studies: compare candidate versions **within**
+each project. Their GCP showcase tests require prepared native runtimes and rich
+capture observers. Default offline test success does not establish live compatibility.
+
+A [live local OpenKritt + Codex example](https://github.com/guybass/agent-eval-flow/blob/main/docs/implementation/local_openkritt.md)
+has now completed against 22 real Flaskr files: eight workflow executions, four
+post-processing executions, 20 tool results and two candidate findings. It saves
+native evidence, verifies the result round trip, and regrades without new model
+calls. Its score checks integration behavior, not security accuracy.
+
+The separate [local OpenSRE + Codex example](https://github.com/guybass/agent-eval-flow/blob/main/docs/implementation/local_opensre.md)
+has completed an investigation of the downloaded HDFS sample: six ReAct
+iterations, eight real tool calls and a cited incident report. It retains native
+events and CLI receipts, then saves, reloads and regrades the captured result.
+
+## Verification and design
+
+Release checks cover the offline suite, archived-run import and regrading,
+package contents, and installed-wheel reporting. See the
+[developer-preview release notes](https://github.com/guybass/agent-eval-flow/blob/main/docs/RELEASE.md) for the verified release scope.
+
+CI runs on **Windows and Ubuntu with Python 3.11, 3.12 and 3.13**. Checks cover
+native-format fixtures, evidence transport, missing outcomes, persistence,
+regrading and package contents. All **90 frozen acceptance/fixture files** are
+verified by hash. Live model tests are opt-in; offline success does not establish
+compatibility with every deployed agent runtime. Historical checkpoints remain
+in [implementation status](https://github.com/guybass/agent-eval-flow/blob/main/docs/IMPLEMENTATION.md).
+
+- [Data structures and configuration](https://github.com/guybass/agent-eval-flow/blob/main/docs/DATA_CONTRACT.md)
+- [Unified assessment flow: configuration and execution in parallel](https://github.com/guybass/agent-eval-flow/blob/main/docs/UNIFIED_ASSESSMENT_FLOW.md)
+- [Module tree and communication graph](https://github.com/guybass/agent-eval-flow/blob/main/docs/lld/README.md)
+- [Real-workflow E2E scenarios](https://github.com/guybass/agent-eval-flow/blob/main/tests/e2e/SHOWCASE.md)
+- [Visual pipeline and score examples](https://github.com/guybass/agent-eval-flow/blob/main/docs/AGENT_EVAL_FLOW_VISION.md)
+- [Library reuse decisions](https://github.com/guybass/agent-eval-flow/blob/main/docs/INTEGRATION_DECISION.md)
+- [What this adds to existing work](https://github.com/guybass/agent-eval-flow/blob/main/docs/WHAT_AGENT_EVAL_FLOW_ADDS.md)
+
+The older design and test-writing checkpoints are retained as historical records.
+
+For changes and bug reports, see [CONTRIBUTING.md](https://github.com/guybass/agent-eval-flow/blob/main/CONTRIBUTING.md).