bundle-run

name	bundle-run
description	Take a completed simulation run and produce a shareable artifact — a blog post, a PDF, or both. Use when the user wants to "publish this forecast", "bundle a run", "make a blog post from this simulation", "generate a PDF report", or "share this forecast". Strips internal-only fields, formats horizons as concrete dates, and pipes through existing publishing flows where available.
allowed-tools	Bash, Read, Write, Edit, Glob

Bundle a run for sharing

Turn a reports/<timestamp>/ directory into something a human reader (or blog audience) can consume.

Inputs to gather

1. Run dir — path to a completed run. Defaults to the latest under the current workspace's reports/. If the runstore is configured, the user can also pick from <runstore_root>/by-date/.
2. Format — blog, pdf, or both. Default: ask.
3. Audience — public (sanitised, neutral) or internal (preserves model names, raw confidence scores, divergence detail). Default: public for blog, internal for pdf-only.
4. Destination blog (if format includes blog) — blog.danielrosehill.com or heyitworks.tech or other configured target. Hand off to daniel-blog-manager:new-blog-post if available.

Steps

1. Locate sentinel. Read .template-id from the workspace containing the run. This determines which output files exist (report.pdf, synthesis.json, chairman_report.md, etc. differ by template).
2. Read run artifacts:
   • Council template: meta.json, head.json, synthesis.json, report.pdf.
   • Forecaster template: meta.json, chairman_report.md, intel_report.pdf, simulation.json.
3. Compute horizon → absolute-date mapping from meta.json's run timestamp + horizon strings. Replace relative horizons ("1w", "1m") with concrete UTC dates in the output.
4. For blog format:
   • Generate Markdown with frontmatter (title, date, tags, excerpt). Title default: Forecast: <question> — <date>.
   • Body: SITREP / executive summary, then the consensus predictions per horizon, then the divergence callouts. Embed the headline visualisations (confidence bars, tone classification) as images linked from the run dir.
   • For public audience: strip per-model attribution, raw confidence floats, and any internal-only metadata. For internal: keep them.
   • Hand off to daniel-blog-manager:new-blog-post if the user picked a blog destination.
5. For pdf format:
   • Use the run's existing report.pdf / intel_report.pdf as the base — the upstream templates already render polished PDFs. Don't re-render unless the audience demands sanitisation.
   • For public sanitisation, regenerate via Typst (lean on repo-to-content:repo-to-pdf patterns) with model names redacted and brand-styled cover.
6. Output paths. Write outputs into <run-dir>/bundled/ so they live alongside the run rather than scattered. Symlink from the runstore's by-date view if configured.
7. Report back. Paths to generated artifacts, blog post URL if published, suggested next steps (e.g. add to Forecasts-Index repo if Daniel maintains one).

Notes

• The Forecaster template's chairman_report.md is already publication-quality prose — for blog format prefer adapting it over rewriting from synthesis.json.
• The Council template's report.pdf is the canonical PDF — bundle-run should default to using it as-is for internal audience.
• Never embed API keys or .env values in the output. Sanity-check the bundled markdown for stray secrets before publishing.