Honeypot Med exports formats that humans and tools can both use: visual proof dossiers, offline proof PDFs, generated UI mockups, GitHub Actions, SARIF, JUnit XML, GitHub step summaries, OpenTelemetry-style logs, OpenInference traces, LangSmith JSONL, promptfoo, Inspect AI, Hugging Face-ready cards, canonical JSONL, Markdown, SVG badges, casebooks, and standalone HTML reports.
The root action.yml installs the project, runs challenge mode, appends github-summary.md to the workflow summary, uploads artifacts, and can upload SARIF to GitHub Code Scanning.
- uses: ByteWorthyLLC/honeypot-med@main
with:
pack: healthcare-challenge
output-dir: honeypot-med-report
fail-under: 70
upload-artifact: true
upload-sarif: true
Each format exists because a different audience needs to consume the same result.
Human-readable proof page for buyers, leadership, screenshots, and public galleries.
Print-friendly HTML proof surface for non-terminal review.
No-API proof document generated locally for attachments and handoffs.
Static product surface generated from the run so stakeholders can see what the result means.
Static-analysis format that can be uploaded into GitHub Code Scanning.
CI-friendly test report where each trap is a testcase and risky traps become failures.
Copy-ready step summary for GitHub Actions runs.
Operational log payload that maps report events into telemetry-style records.
Offline JSONL trace shapes for teams that already inspect LLM runs in observability tools.
Offline adapters for promptfoo, Inspect AI, legacy OpenAI Evals, and canonical JSONL workflows.
Dataset card, system card, and artifact manifest generated without upload or API calls.
Forensic HTML, traparium, unknowns page, failure recipes, trap tree, and notebook.
Raster social card and badge generated with the Python standard library.
Zip archive, SHA-256 checksum file, manifest, and release notes for generated report directories.
Stable machine-readable report for custom dashboards and downstream eval systems.
PR comments, release notes, docs, and simple diffs.
README, docs, and launch posts that show the result without opening the report.
The default generated promptfoo config uses the echo provider, so adapter generation and assertion checks can stay offline until a team intentionally swaps in a live provider. The verify command checks that the generated adapter directory is parseable.
python app.py eval-kit --pack healthcare-challenge --outdir reports/eval-kit --engine-mode local --no-allow-network
python app.py eval-kit verify --dir reports/eval-kit python app.py hf-mirror plan --outdir reports/hf-mirror
python app.py release-kit --source-dir reports/export --outdir dist/release-bundles --name claims-export