Movement Leader Harvest Manifest

Movement Leader Harvest Manifest Builder

Produce a correct, ingest-ready harvest-manifest.json for one tenant from whatever files and catalogs already exist. This skill is the manifest factory — it does not sweep the web (see movement-leader-content-harvest) and does not upload to Supabase (see movement-leader-harvest-ingest).

Invocation

/movement-leader-harvest-manifest {leader-slug}
/movement-leader-harvest-manifest danielle-strickland --from-harvest-only
/movement-leader-harvest-manifest rowland-smith --merge-existing --validate
/movement-leader-harvest-manifest hugh-halter --supplement links.csv

Run from movement-leader-websites/ monorepo root unless cwd is inside a leader repo.

CLI (preferred first pass)

pnpm harvest:build-manifest -- --slug {leader-slug}
pnpm harvest:build-manifest -- --slug danielle-strickland --dry-run
pnpm harvest:build-manifest -- --slug rowland-smith --merge-existing
pnpm harvest:build-manifest -- --slug hugh-halter --supplement docs/movement_leader_research/content/link-only.json
pnpm harvest:validate-manifest -- --slug danielle-strickland
pnpm harvest:build-bundle -- --repo ./danielle-strickland --out ./danielle-strickland-bundle.zip

Resolve the tenant

1. Normalize user input to kebab-case {slug}.
2. Confirm {MONOREPO_ROOT}/{slug}/ exists (or cwd is that repo).
3. Set paths:

Exclude: _template, _template-full, docs, scripts, shared, reports.

Workflow (execute in order)

Copy this checklist and mark steps as you go:

- [ ] 1. Inventory sources
- [ ] 2. Classify each item (type + disposition)
- [ ] 3. Assign stable ids + harvested_path
- [ ] 4. Merge LINK-only / catalog rows
- [ ] 5. Write manifest + summary
- [ ] 6. Validate (CLI + ingest preview)

Step 1 — Inventory sources

Scan in priority order (do not double-count):

1. docs/harvest/{type}/*.md — primary; frontmatter is authoritative when present
2. docs/harvest/transcripts/*.md — attach to existing media rows; do not duplicate as articles
3. Legacy folders (if --merge-existing or no harvest dir): docs/articles/, docs/videos/, docs/podcasts/
4. Existing manifest — preserve LINK/TRANSCRIBE/SKIP rows when merging
5. Supplement file — CSV/JSON for platform-only media (YouTube, podcast feeds)

Run: pnpm harvest:build-manifest -- --slug {slug} --dry-run and read the printed source counts.

Step 2 — Classify type and disposition

Type comes from (first match wins):

1. Frontmatter type: field
2. Parent directory under docs/harvest/ (see reference/type-taxonomy.md)
3. Heuristics on source_url / file path (youtube → video, substack → newsletter)

Disposition (see rubric in reference):

Never put books in the manifest.

Step 3 — Stable ids and paths

Id format: {slug}-{type-short}-{NNN} or reuse frontmatter id / filename stem.

Examples: danielle-strickland-blog-094, jr-woodward-newsletter-003, hugh-halter-video-012.

harvested_path (FETCH only): repo-relative POSIX path:

docs/harvest/{type-folder}/{id}.md

Type folder names (plural): articles, newsletters, videos, podcasts, interviews, academic, talks, audio.

Frontmatter template for new FETCH files — see reference/frontmatter-template.md.

Step 4 — Merge LINK-only rows

Videos and podcast-host episodes often have no markdown file. Add them via:

• --merge-existing (keep prior LINK rows), and/or
• --supplement JSON:

[
  {
    "id": "hugh-halter-video-001",
    "title": "Talk title",
    "type": "video",
    "disposition": "LINK",
    "canonical_url": "https://www.youtube.com/watch?v=abc123",
    "host": "YouTube",
    "date": "2021-06-01",
    "primary": true
  }
]

Dedupe by id first, then by normalized canonical_url.

Step 5 — Write manifest

Output schema: reference/manifest-schema.md.

pnpm harvest:build-manifest -- --slug {slug} --merge-existing

Also refresh CONTENT_SOURCES.md stub summary if the tenant already uses it (counts must match non-SKIP items).

Step 6 — Validate

pnpm harvest:validate-manifest -- --slug {slug}

Gate checks:

• Every FETCH row has harvested_path pointing to an existing file
• Every LINK video/podcast row has canonical_url
• No duplicate id values
• summary.total == items.length
• Summary disposition counts match items
• No type: book in items
• leader field matches slug

ZenWrite admin bundle: run from zenwrite repo:

pnpm harvest:build-bundle -- --repo ../movement-leader-websites/{slug} --out {slug}-bundle.zip

Upload zip in Admin → Harvest import → Preview must show correct targetTable per row.

Type → ingest target (downstream)

Full mapping: reference/type-taxonomy.md.

Agent execution rules

1. Prefer the CLI over hand-editing JSON — it keeps paths and summaries consistent.
2. Read frontmatter before inferring — harvested files often already contain full metadata.
3. Do not invent URLs or dates — use unknown for date when missing.
4. Migration, not authoring — manifest describes existing content only.
5. Persist incrementally — write manifest after each batch when fixing large tenants.
6. No app code changes unless adding manifest tooling under scripts/harvest/.

Session report (print when done)

1. {slug} and manifest path
2. Items by type and disposition (FETCH / LINK / TRANSCRIBE / SKIP)
3. Validation: pass / fail with first 5 errors
4. fetched_to_disk count vs files found on disk
5. Suggested next step: pnpm harvest:transcribe → pnpm harvest:ingest or ZenWrite bundle upload

Related skills

Additional resources

• reference/manifest-schema.md
• reference/type-taxonomy.md
• reference/frontmatter-template.md
• Step 08 prompt: docs/build/skills/content-gathering/08-write-inventory-and-manifest.md
• ZenWrite ingest mapper: zenwrite/server/services/admin/ingest/harvestMapper.ts