agent-release-gates 0.1.0__tar.gz

agent_release_gates-0.1.0/.dockerignore

@@ -0,0 +1,22 @@
+.git/
+.github/
+.venv/
+__pycache__/
+*.py[cod]
+.pytest_cache/
+.ruff_cache/
+.coverage
+htmlcov/
+.env
+.env.*
+!.env.example
+dist/
+build/
+*.egg-info/
+internal_ai_agent_project_plan.md
+reports/collector_export_smoke.json
+reports/collector_deployment_check.json
+reports/provider_embedding_eval_status.json
+reports/provider_embedding_eval_summary.json
+reports/provider_embedding_eval_cases.jsonl
+public/

agent_release_gates-0.1.0/.gitattributes

@@ -0,0 +1 @@
+*.pdf binary

agent_release_gates-0.1.0/.github/ISSUE_TEMPLATE/bug_report.yml

@@ -0,0 +1,46 @@
+name: Bug or reproducibility issue
+description: Report a failing command, broken artifact, dashboard issue, or reproducibility problem.
+title: "[Bug]: "
+labels: ["bug"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Please include enough context to reproduce the issue. Do not include API keys, credentials, private data, production logs, or confidential examples.
+  - type: textarea
+    id: what_happened
+    attributes:
+      label: What happened?
+      description: Describe the failure or unexpected behavior.
+    validations:
+      required: true
+  - type: textarea
+    id: command
+    attributes:
+      label: Command or page
+      description: Paste the command, URL, or page where the issue occurred. Redact secrets.
+      placeholder: "uv run python scripts/run_all_evals.py"
+    validations:
+      required: false
+  - type: textarea
+    id: expected
+    attributes:
+      label: Expected behavior
+      description: What did you expect to happen?
+    validations:
+      required: true
+  - type: textarea
+    id: environment
+    attributes:
+      label: Environment
+      description: OS, Python version, browser, or relevant setup detail.
+      placeholder: "Windows 11, Python 3.12, uv installed"
+    validations:
+      required: false
+  - type: checkboxes
+    id: checks
+    attributes:
+      label: Safety check
+      options:
+        - label: I have removed credentials, personal data, private company data, and confidential logs.
+          required: true

agent_release_gates-0.1.0/.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,11 @@
+blank_issues_enabled: false
+contact_links:
+  - name: Reviewer handoff pack
+    url: https://rosscyking1115.github.io/agent-release-gates/reviewer_handoff_pack.md
+    about: Read the external-review instructions before volunteering to label cases.
+  - name: Public dashboard
+    url: https://agent-evaluation-lab.streamlit.app/
+    about: Explore the interactive evaluation dashboard.
+  - name: Full public report
+    url: https://rosscyking1115.github.io/agent-release-gates/evaluation_report.html
+    about: Review the generated evaluation report and current limitations.

agent_release_gates-0.1.0/.github/ISSUE_TEMPLATE/eval_improvement.yml

@@ -0,0 +1,51 @@
+name: Benchmark or evaluation improvement
+description: Suggest a dataset, metric, failure mode, intervention, or documentation improvement.
+title: "[Eval improvement]: "
+labels: ["evaluation", "enhancement"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Use this for benchmark-quality improvements. Please keep the project boundary conservative: no private company documents, credentials, production logs, customer data, employee data, or confidential workflows.
+  - type: dropdown
+    id: area
+    attributes:
+      label: Area
+      options:
+        - Retrieval or RAG grounding
+        - Safety classifier or prevalence estimation
+        - Agent/tool governance
+        - Memory/context pollution
+        - Goal-conflict arbitration
+        - Multi-model comparison
+        - Human review or judge reliability
+        - Public report or dashboard
+        - Documentation
+        - Other
+    validations:
+      required: true
+  - type: textarea
+    id: proposal
+    attributes:
+      label: Proposal
+      description: What should change, and why would it improve evaluation quality?
+      placeholder: "Describe the failure mode, metric, dataset, or report improvement."
+    validations:
+      required: true
+  - type: textarea
+    id: reproducibility
+    attributes:
+      label: Reproducibility plan
+      description: How could the change be tested or regenerated?
+      placeholder: "Example: add cases, update expected metrics, run scripts/run_all_evals.py and pytest."
+    validations:
+      required: false
+  - type: checkboxes
+    id: data_boundary
+    attributes:
+      label: Data boundary
+      options:
+        - label: This suggestion does not require private, confidential, personal, or production data.
+          required: true
+        - label: If using a public dataset, I can provide source, license, and sampling details.
+          required: false

agent_release_gates-0.1.0/.github/ISSUE_TEMPLATE/external_review.yml

@@ -0,0 +1,47 @@
+name: External review volunteer
+description: Volunteer to independently review calibration cases or benchmark claims.
+title: "[External review]: "
+labels: ["external-review"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Thank you for helping review Agent Release Safety Gates.
+
+        Please do not paste completed labels, personal information, credentials, private documents, or confidential examples into a public issue. This issue is only for volunteering, clarifying scope, or reporting review-process questions.
+  - type: dropdown
+    id: review_type
+    attributes:
+      label: Review type
+      options:
+        - Safety-label calibration cases
+        - Report or benchmark-methodology review
+        - Dataset-boundary review
+        - Reproducibility review
+        - Other review
+    validations:
+      required: true
+  - type: checkboxes
+    id: independence
+    attributes:
+      label: Independence check
+      options:
+        - label: I have not inspected maintainer labels, classifier outputs, hosted judge outputs, or source code for the cases I plan to label.
+          required: false
+        - label: I can use a pseudonymous reviewer id and avoid sharing personal information in label files.
+          required: true
+  - type: textarea
+    id: context
+    attributes:
+      label: Relevant context
+      description: Briefly describe your review background or what you want to review. Do not include private data.
+      placeholder: "Example: I can independently label the 24 safety calibration cases this week."
+    validations:
+      required: true
+  - type: textarea
+    id: questions
+    attributes:
+      label: Questions or constraints
+      description: Add any process questions or availability constraints.
+    validations:
+      required: false

agent_release_gates-0.1.0/.github/pull_request_template.md

@@ -0,0 +1,17 @@
+## Summary
+
+Describe the evaluation, documentation, dashboard, or reproducibility change.
+
+## Checks
+
+- [ ] Synthetic and public-data tracks remain separated.
+- [ ] No private company documents, customer data, employee data, credentials, production logs, or confidential workflows were added.
+- [ ] New or changed metrics include limitations.
+- [ ] Generated reports or public-site artifacts were regenerated when needed.
+- [ ] Tests or checks were run, or the reason they were skipped is documented.
+
+## Verification
+
+```text
+Paste relevant commands and results here.
+```

agent_release_gates-0.1.0/.github/workflows/ci.yml

@@ -0,0 +1,53 @@
+name: ci
+
+on:
+  push:
+    branches: ["main"]
+  pull_request:
+  workflow_dispatch:
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check out repository
+        uses: actions/checkout@v6
+
+      - name: Set up Python
+        uses: actions/setup-python@v6
+        with:
+          python-version: "3.12"
+
+      - name: Install uv
+        run: python -m pip install uv
+
+      - name: Install dependencies
+        run: uv sync --locked --dev
+
+      - name: Lint
+        run: uv run ruff check .
+
+      - name: Test
+        run: uv run pytest
+
+      - name: Regenerate deterministic eval reports
+        run: uv run python scripts/run_all_evals.py
+
+      - name: Smoke OTLP collector export
+        run: uv run python scripts/smoke_otel_collector.py
+
+      - name: Dry-run provider embedding eval
+        run: uv run python scripts/run_provider_embedding_eval.py
+
+      - name: Start OpenTelemetry Collector
+        run: docker compose --profile observability up -d otel-collector
+
+      - name: Check OpenTelemetry Collector deployment
+        run: uv run python scripts/check_otel_collector_deployment.py
+
+      - name: Stop OpenTelemetry Collector
+        if: always()
+        run: docker compose --profile observability down
+
+      - name: Build Docker image
+        run: docker build -t agent-release-gates:ci .

agent_release_gates-0.1.0/.github/workflows/pages.yml

@@ -0,0 +1,59 @@
+name: pages
+
+on:
+  push:
+    branches: ["main"]
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  pages: write
+  id-token: write
+
+env:
+  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: true
+
+concurrency:
+  group: "pages"
+  cancel-in-progress: false
+
+jobs:
+  deploy:
+    environment:
+      name: github-pages
+      url: ${{ steps.deployment.outputs.page_url }}
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check out repository
+        uses: actions/checkout@v6
+
+      - name: Set up Python
+        uses: actions/setup-python@v6
+        with:
+          python-version: "3.12"
+
+      - name: Install uv
+        run: python -m pip install uv
+
+      - name: Install dependencies
+        run: uv sync --locked --dev
+
+      - name: Regenerate deterministic eval reports
+        run: uv run python scripts/run_all_evals.py
+
+      - name: Build public site
+        run: uv run python scripts/build_public_site.py
+
+      - name: Configure Pages
+        uses: actions/configure-pages@v5
+        with:
+          enablement: true
+
+      - name: Upload Pages artifact
+        uses: actions/upload-pages-artifact@v3
+        with:
+          path: public
+
+      - name: Deploy to GitHub Pages
+        id: deployment
+        uses: actions/deploy-pages@v4

agent_release_gates-0.1.0/.gitignore

@@ -0,0 +1,25 @@
+.venv/
+__pycache__/
+*.py[cod]
+.pytest_cache/
+.ruff_cache/
+.coverage
+htmlcov/
+.env
+.env.*
+!.env.example
+dist/
+build/
+*.egg-info/
+reports/collector_export_smoke.json
+reports/collector_deployment_check.json
+reports/provider_embedding_eval_status.json
+reports/provider_embedding_eval_summary.json
+reports/provider_embedding_eval_cases.jsonl
+reports/model_judge_eval_status.json
+reports/model_judge_eval_summary.json
+reports/model_judge_eval_cases.jsonl
+data/public/techqa_train.json
+data/public/wixqa_expertwritten_test.jsonl
+data/public/wixqa_kb_corpus.jsonl
+/public/

agent_release_gates-0.1.0/.streamlit/config.toml

@@ -0,0 +1,2 @@
+[browser]
+gatherUsageStats = false

agent_release_gates-0.1.0/CONTRIBUTING.md

@@ -0,0 +1,69 @@
+# Contributing
+
+Thanks for taking a look at Agent Release Safety Gates.
+
+This project is intended to be a public benchmark-style artifact. Contributions should improve evaluation quality, reproducibility, safety analysis, or clarity. Please keep the data boundary conservative: do not add private company documents, customer data, employee data, credentials, production logs, or confidential workflows.
+
+## Good Contribution Areas
+
+- hand-authored benign, ambiguous, unsafe, prompt-injection, weak-evidence, excessive-agency, and tool-misuse cases
+- clearer failure taxonomy labels
+- public benchmark expansion with reproducible sampling
+- judge-reliability experiments
+- human-review calibration workflow
+- multi-model adapters with explicit model ids, dates, settings, and limitations
+- report, dashboard, or documentation improvements that make results easier to inspect
+- tests that prevent benchmark or reporting regressions
+
+## Opening Issues
+
+Use GitHub issues for public, non-sensitive project discussion:
+
+- External review volunteer: use this to offer independent labels, methodology review, reproducibility review, or dataset-boundary review.
+- Benchmark or evaluation improvement: use this for proposed datasets, metrics, failure modes, interventions, or report improvements.
+- Bug or reproducibility issue: use this for failing commands, broken artifacts, dashboard problems, or unclear setup.
+
+Do not paste completed reviewer labels, credentials, personal data, private company data, production logs, or confidential examples into public issues.
+
+## Data Rules
+
+- Use synthetic internal-operations data unless a public dataset is explicitly documented.
+- Keep public datasets separated from the synthetic benchmark in code, docs, and reports.
+- Do not add real personal data or confidential business procedures.
+- Do not imply the project measures any real organization's internal AI system.
+
+## Result Rules
+
+- Do not publish provider-backed or multi-model results unless the run is reproducible and the model settings are documented.
+- Separate synthetic labels, simulated reviewer labels, real human labels, and LLM-as-judge labels.
+- Include limitations when adding new metrics.
+- Add or update tests when changing scoring, reports, or release gates.
+
+## Local Checks
+
+Run these before opening a pull request:
+
+```powershell
+uv run ruff check .
+uv run pytest
+uv run python scripts/run_all_evals.py
+uv run python scripts/build_public_site.py
+```
+
+Optional checks:
+
+```powershell
+uv run python scripts/smoke_otel_collector.py
+docker compose --profile observability up -d otel-collector
+uv run python scripts/check_otel_collector_deployment.py
+docker compose --profile observability down
+docker build -t agent-release-gates:local .
+```
+
+## Pull Request Checklist
+
+- The change has a clear evaluation or documentation purpose.
+- Synthetic and public-data tracks remain separated.
+- New metrics are reproducible.
+- New limitations are documented.
+- Tests or generated artifacts are updated when needed.

agent_release_gates-0.1.0/Dockerfile

@@ -0,0 +1,34 @@
+FROM ghcr.io/astral-sh/uv:python3.12-bookworm-slim
+
+WORKDIR /app
+
+ENV PYTHONDONTWRITEBYTECODE=1
+ENV PYTHONUNBUFFERED=1
+ENV UV_LINK_MODE=copy
+ENV STREAMLIT_BROWSER_GATHER_USAGE_STATS=false
+# Sync the env once at build time; runtime `uv run` uses it without re-syncing
+# (which would otherwise drop the optional api/dashboard extras).
+ENV UV_NO_SYNC=1
+
+COPY pyproject.toml uv.lock README.md streamlit_app.py ./
+COPY src ./src
+
+RUN uv sync --locked --no-dev --extra api --extra dashboard
+
+COPY app ./app
+COPY data ./data
+COPY docs ./docs
+COPY reports ./reports
+COPY scripts ./scripts
+COPY .streamlit ./.streamlit
+
+RUN uv run --no-dev python scripts/run_all_evals.py
+
+RUN useradd --create-home --shell /usr/sbin/nologin appuser \
+    && chown -R appuser:appuser /app
+
+USER appuser
+
+EXPOSE 8000
+
+CMD ["uv", "run", "--no-dev", "uvicorn", "internal_ai_agent.api.main:app", "--host", "0.0.0.0", "--port", "8000"]

agent_release_gates-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 rosscyking1115
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

agent_release_gates-0.1.0/PKG-INFO

@@ -0,0 +1,201 @@
+Metadata-Version: 2.4
+Name: agent-release-gates
+Version: 0.1.0
+Summary: Release-readiness gates for AI agents: replay known incidents, apply policy-as-code gates, and produce ship/warn/block evidence before an agent, prompt, model, or tool-policy change ships.
+Project-URL: Homepage, https://github.com/rosscyking1115/agent-release-gates
+Project-URL: Repository, https://github.com/rosscyking1115/agent-release-gates
+Project-URL: Documentation, https://rosscyking1115.github.io/agent-release-gates/
+Project-URL: Issues, https://github.com/rosscyking1115/agent-release-gates/issues
+Author: rosscyking1115
+License-Expression: MIT
+License-File: LICENSE
+Keywords: agent,ai-safety,evaluation,inspect-ai,llm,red-team,release-gate
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Software Development :: Quality Assurance
+Requires-Python: >=3.12
+Requires-Dist: pydantic>=2.8.0
+Provides-Extra: api
+Requires-Dist: fastapi>=0.115.0; extra == 'api'
+Requires-Dist: uvicorn[standard]>=0.30.0; extra == 'api'
+Provides-Extra: dashboard
+Requires-Dist: altair>=6.0.0; extra == 'dashboard'
+Requires-Dist: pandas>=2.2.0; extra == 'dashboard'
+Requires-Dist: streamlit>=1.38.0; extra == 'dashboard'
+Description-Content-Type: text/markdown
+
+# Agent Release Safety Gates
+
+A public release-readiness system for testing whether AI-agent workflow changes remain grounded, safe, auditable, and useful under retrieval, refusal, prompt-injection, incident replay, and approval-gated tool-use conditions.
+
+This project is not a clone, assessment, or reverse-engineering attempt of any company's internal AI system. The controlled operations benchmark is synthetic by design; TechQA and WixQA are used separately as public retrieval-validation datasets.
+
+## Live Project
+
+- Public project page: https://rosscyking1115.github.io/agent-release-gates/
+- Full evaluation report (HTML): https://rosscyking1115.github.io/agent-release-gates/evaluation_report.html
+- PDF report: https://rosscyking1115.github.io/agent-release-gates/evaluation_report.pdf
+- Interactive dashboard: run locally (see [Run Locally](#run-locally)) or deploy on
+  Streamlit Cloud — see the [dashboard deployment guide](docs/dashboard.md). A hosted
+  instance must be set to public visibility to be shareable.
+
+## What This Project Does
+
+The project evaluates an AI-agent workflow across five release questions:
+
+- Does the agent retrieve the right evidence and cite it?
+- Does it abstain or refuse when evidence is weak, unsafe, or prompt-injected?
+- Does it require approval before mock side-effecting tool calls?
+- Does it leave enough trace, audit, and monitoring evidence for review?
+- Does it pass incident replay and policy-as-code release gates?
+
+The result is a reproducible evaluation artifact rather than a one-off dashboard: deterministic eval runners, generated reports, CI checks, Dockerized local execution, a Streamlit dashboard, and a GitHub Pages report site.
+
+## Product Direction
+
+**Agent Release Safety Gates** is a release-readiness workflow for replaying known agent incidents, applying policy gates, and producing evidence before a changed agent, prompt, model, or tool policy ships. Its deterministic evaluation benchmark and runners are the evidence layer behind that workflow.
+
+The first module is an Incident Replay Suite that turns redacted synthetic incidents into regression fixtures, replay results, release gates, and incident memos.
+
+To evaluate an external agent, use the public quickstart:
+
+- [Evaluate your agent quickstart](docs/evaluate_your_agent_quickstart.md)
+- [Incident pack schema](docs/incident_pack_schema.md)
+- [Candidate results schema](docs/candidate_results_schema.md)
+
+## Current Evidence Snapshot
+
+| Area | Current result |
+| --- | --- |
+| Controlled benchmark | 358 synthetic golden cases, 60 red-team cases, 180 synthetic operations tickets |
+| Retrieval | 100.00% synthetic retrieval hit rate@3 with local TF-IDF/vector-style retrievers |
+| Public RAG validation | 160 TechQA cases and 80 WixQA cases evaluated separately from the synthetic benchmark |
+| Safety | 90.91% classifier recall, 0 high-severity false negatives in the current challenge set |
+| Agent governance | 100.00% mock side-effect block rate and approval audit rate |
+| Incident replay | 8 seeded synthetic incidents replayed, 100.00% closure rate, 0 replay must-not violations |
+| Intervention study | 3 deterministic safety studies plus public RAG grounding and memory/context studies |
+| Hosted judge calibration | Reviewed OpenAI and Anthropic judge runs with public-safe provider comparison |
+
+These results are engineering evidence over controlled benchmarks. They are not claims of real-world production performance.
+
+## Key Findings
+
+- Safety metrics are not meaningful alone; the lab reports over-review cost, benign auto-blocks, weak-evidence handling, and unsafe misses beside the headline scores.
+- Layered safeguards reduce selected prompt-injection, unsafe-action, and unsafe-request failures in controlled studies while making review burden visible.
+- Public TechQA and WixQA retrieval tracks help test whether the RAG harness works beyond self-contained synthetic data.
+- Public RAG grounding thresholds reduce unsupported answer attempts while making abstention and review cost visible.
+- Memory/context controls reduce polluted-memory following while preserving benign memory usefulness.
+- Goal-conflict arbitration reduces unsafe goal-following while preserving benign task completion.
+- Synthetic operations data remains useful for controlled tests that would be unsafe or impractical to run on confidential real workflows.
+- The next strongest validation step is independent human labelling, followed by broader multi-model comparison.
+
+## What Is Included
+
+- Evaluation runners for retrieval, extraction, safety classification, controlled-agent behavior, and observability.
+- Baseline-vs-intervention studies for instruction hierarchy, action-risk gates, and safety classifier review policy.
+- Public RAG grounding and abstention intervention study over TechQA and WixQA.
+- Memory/context pollution intervention study covering stale, injected, and cross-user memory.
+- Goal-conflict intervention study covering safety, evidence, privacy, and tool-risk arbitration.
+- Incident replay suite with seeded incidents, replay matrix, release gates, regression fixtures, and generated memos.
+- Public benchmark documentation, dataset boundaries, failure taxonomy, and external-review packet.
+- Candidate-results exporters for generic agent logs and LangChain/LangSmith-style traces.
+- Streamlit dashboard for interactive inspection.
+- GitHub Pages report and PDF for public review.
+- CI, Docker, Docker Compose, linting, tests, and deterministic report regeneration.
+
+## Install
+
+Once published to PyPI (see [publishing guide](docs/publishing.md)), the core install
+is lean — `pydantic` only — and gives you the CLI, the Inspect suite, the real-agent
+runner, and the scoring logic. The API and dashboard are opt-in extras:
+
+```bash
+pip install agent-release-gates                # CLI + Inspect suite + scoring
+pip install "agent-release-gates[api]"         # + FastAPI evidence service
+pip install "agent-release-gates[dashboard]"   # + Streamlit dashboard deps
+pip install agent-release-gates inspect_ai     # to run under Inspect
+```
+
+```bash
+agent-safety release-gate                                          # ship / warn / block
+inspect eval agent-release-gates/incident_replay --model openai/gpt-4.1-mini
+```
+
+> Not yet on PyPI — build it yourself with `uv build`, or run from source below.
+
+## Run Locally
+
+```powershell
+uv sync
+uv run python scripts/run_all_evals.py
+# Release gate (installed console command); exits non-zero on a blocking failure.
+uv run agent-safety release-gate --policy config/incident_release_policy.json
+# Interactive dashboard.
+uv run streamlit run streamlit_app.py --server.port 8510
+```
+
+Open `http://localhost:8510`. Run the API and dashboard together with
+`docker compose up --build`, then open `http://localhost:8510` and
+`http://localhost:8000/health`.
+
+Drive a real LLM through the release gate, or run the suite under Inspect:
+
+```powershell
+# Any OpenAI-compatible / self-hosted open model endpoint.
+$env:AGENT_RUNNER_API_KEY = "..."
+uv run python scripts/run_real_agent_replay.py
+
+# Inspect (UK AISI) -- optional peer dependency.
+uv pip install inspect_ai
+inspect eval agent-release-gates/incident_replay --model openai/gpt-4.1-mini
+```
+
+## Verification
+
+```powershell
+uv run ruff check .
+uv run pytest
+uv run python scripts/run_all_evals.py
+uv run agent-safety release-gate --policy config/incident_release_policy.json
+uv run python scripts/build_public_site.py
+docker build -t agent-release-safety-gates:local .
+```
+
+CI runs linting, tests, deterministic report checks, local OpenTelemetry smoke testing, Dockerized collector verification, and Docker build verification.
+
+## Review Materials
+
+- Evaluate your agent quickstart: [docs/evaluate_your_agent_quickstart.md](docs/evaluate_your_agent_quickstart.md)
+- Benchmark card: [docs/benchmark_card.md](docs/benchmark_card.md)
+- Agent safety intervention study: [docs/agent_safety_intervention_study.md](docs/agent_safety_intervention_study.md)
+- RAG grounding intervention report: [reports/rag_grounding_intervention.md](reports/rag_grounding_intervention.md)
+- Memory context intervention report: [reports/memory_context_intervention.md](reports/memory_context_intervention.md)
+- Goal conflict intervention report: [reports/goal_conflict_intervention.md](reports/goal_conflict_intervention.md)
+- Incident pack schema: [docs/incident_pack_schema.md](docs/incident_pack_schema.md)
+- Candidate results schema: [docs/candidate_results_schema.md](docs/candidate_results_schema.md)
+- Incident replay summary: [reports/incident_replay_summary.json](reports/incident_replay_summary.json)
+- Dataset card: [docs/dataset_card.md](docs/dataset_card.md)
+- Failure taxonomy: [docs/failure_taxonomy.md](docs/failure_taxonomy.md)
+- External reviewer handoff pack: [docs/reviewer_handoff_pack.md](docs/reviewer_handoff_pack.md)
+- Technical artifact index: [docs/technical_artifacts.md](docs/technical_artifacts.md)
+- Contribution guide: [CONTRIBUTING.md](CONTRIBUTING.md)
+
+## Current Limitations
+
+- The controlled benchmark is synthetic and still partly templated.
+- Public TechQA and WixQA tracks use compact samples, not the full upstream datasets.
+- Human-review labels are currently simulated workflow labels; independent reviewer labels are prepared but not yet published.
+- Hosted model evidence includes reviewed judge-calibration runs, not a broad multi-model agent comparison.
+- Provider-backed embedding and reranker adapters are prepared, but credentialed hosted results are not claimed until reviewed.
+
+## Roadmap
+
+- Collect independent human labels using the prepared review packet.
+- Add reproducible multi-model comparison across hosted and open-source models.
+- Expand public RAG validation beyond the current compact TechQA and WixQA samples.
+- Add more framework-specific candidate-results exporters for common agent runners.
+- Expand the paper-style intervention report with external reviewer disagreement analysis.
+- Invite external review through issues and contribution guidelines.