fix: give hand-authored toolkit prose a source home in curation/

[teallarson]

Why

Hand-authored toolkit prose was stored only in generated data/toolkits/*.json. A forced regeneration has no previous output to carry forward, so it could silently delete documentationChunks, customImports, and subPages.

This gives that content a durable source of truth in curation/, so nightly generation can rebuild the JSON without losing editorial work.

How it fits

API + metadata ─┐
├─> generator + merger ─> data/toolkits/*.json ─> docs app
curation/ ──────┘ │
└─ force-regenerate remains lossless

The curation layer supplies the fields that do not come from an upstream API or LLM. It is loaded through the existing --custom-sections path and takes precedence over carry-forward from previous output.

Changes

• Added 75 per-toolkit curation files, extracted verbatim from the committed artifacts.
• Updated the custom-sections source to load a directory while retaining single-file compatibility.
• Updated the nightly workflow to pass --custom-sections ./curation.
• Added a re-runnable extraction script and regression coverage for forced regeneration.

Verification

• All 117 toolkits reproduce their committed prose byte-for-byte.
• Spot-checked rendered output across GitHub, Google Flights, Google News, ClickUp, and Jira.
• Generator tests, typecheck, and lint pass.

Independent of #1106 and #1113; #1111 documents the hazard this fixes.

---

Note

Medium Risk
Large docs corpus and generator pipeline changes affect nightly doc output and merge behavior; failures are mostly explicit validation rather than silent data loss.

Overview
Hand-authored toolkit docs (documentationChunks, imports, subpages) now live in toolkit-docs-generator/curation/<toolkit>/ as YAML-frontmatter MDX instead of only in generated JSON. A new MarkdownCurationSource compiles chunks/, imports/, and pages/ into the existing custom-sections shape; the old JSON file loader is removed.

When a curation root is configured (default ./curation if present, and --custom-sections ./curation in the nightly workflow), that tree is authoritative: missing toolkit folders mean empty prose, deleting the last file clears content on the next run, and invalid frontmatter/MDX or bad tool targets fail the build rather than preserving stale JSON. Merge logic stops carry-forward from previous output in that mode, stores curationSourceHash on artifacts, and limits secret-coherence LLM rewrites on curated chunks (warn only).

check-changes and --skip-unchanged now treat curation edits like API changes via getChangedToolkitIdsFromCustomSections. The PR adds the extracted per-toolkit MDX corpus plus regression tests (force-regenerate, corpus parity with committed JSON).

^{Reviewed by Cursor Bugbot for commit 99ec7ab. Bugbot is set up for automated code reviews on this repo. Configure here.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
docs	Ready	Preview	Aug 12, 2026 6:53pm

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 6e382f1. Configure here.}

[sdserranog]

This is nice! I like the curation idea, and thanks for adding the context in the Architecture and README files. It would be good to have a skill or something more in-depth for creating one. The examples will help the agents a lot, but I'd also like a table or document that covers the types of curation files, how each frontmatter key works, and what you can do with them.