Movement Leader Content Harvest

Movement Leader Content Harvest

Find all sources of all content for one movement leader, except books, and pull that content (or a durable pointer) into the leader's website repo. This is migration and gap-closing — it moves existing content in from where it already lives. It does not write new articles or generate content.

The governing question for every item: "If this platform vanished tomorrow, would we still have the content?"

Invocation

/movement-leader-content-harvest {leader-slug}
/movement-leader-content-harvest Hugh Halter
/movement-leader-content-harvest deb-hirsch --from-step 02
/movement-leader-content-harvest hugh-halter --players youtube,podcasts --dry-run
/movement-leader-content-harvest rowland-smith --refresh

Run from the monorepo root (movement-leader-websites/) unless you are already inside a leader repo.

Arguments

Books are out of scope. Log any book to books-seen.md and defer to corpus-upload.

Resolve the tenant

1. If the user gave a slug or name, normalize to kebab-case and confirm {MONOREPO_ROOT}/{slug}/ exists.
2. If cwd is {MONOREPO_ROOT}/{slug}/ (has package.json or docs/movement_leader_research/), use that slug.
3. Exclude non-tenant dirs: _template, _template-full, docs, scripts, shared, reports, node_modules.
4. Set paths for the run:
   • TENANT_ROOT = {MONOREPO_ROOT}/{slug}/
   • TENANT_SLUG = {slug}
   • TRACKER = {TENANT_ROOT}/docs/movement_leader_research/content/HARVEST_TRACKER.md
   • REGISTRY = {MONOREPO_ROOT}/docs/build/skills/content-gathering/CONTENT_REGISTRY.md

If ambiguous, ask once. If no dossier exists at all, step 01 bootstraps identity from one web search.

Execution protocol (run to completion)

Read this skill, then execute steps 00–09 sequentially for TENANT_SLUG. Do not stop between steps unless blocked (missing identity, disambiguation failure, or user --to-step). After every step, update HARVEST_TRACKER.md before continuing.

For each step, read the full step prompt from docs/build/skills/content-gathering/ and follow it exactly, substituting {TENANT_SLUG} and {MONOREPO_ROOT} paths:

Also read docs/build/skills/content-gathering/MASTER_PLAYBOOK.md once at the start (paths relative to monorepo root).

Step 01 tracker bootstrap

If HARVEST_TRACKER.md does not exist:

cp docs/build/skills/content-gathering/HARVEST_TRACKER_TEMPLATE.md \
   {TENANT_ROOT}/docs/movement_leader_research/content/HARVEST_TRACKER.md

Parallel sweeps (steps 02–05)

Steps 02–05 are independent. You may run ≤2–3 in parallel (separate subagents), but each must write _raw-finds.md rows and update the tracker before ending. Never exceed 3 concurrent web sweeps — rate limits apply. Step 06 merges all raw finds.

--dry-run override

Skip file writes in step 07. Still run disposition (06) and write inventory stubs (08) with harvested_path blank or DRY-RUN. Do not mark verified in step 09.

--refresh override

In steps 06–08, reconcile against existing CONTENT_SOURCES.md and harvest-manifest.json: append new finds, mark dead links, do not duplicate items already in the skip set.

Disposition rubric (step 06)

Bias: owned + old/fragile primary text → FETCH; healthy third-party → LINK.

Outputs (per tenant)

All paths under {TENANT_ROOT}/:

Reference implementation: andrew-jones — docs/videos/ and docs/articles/old-blogs/ show target shape.

Discovery tools (preference order)

1. WebSearch — channels, archives, author pages, guest appearances
2. WebFetch — enumerate channel/feed/archive pages
3. Bash — RSS/sitemap via curl; yt-dlp if available for YouTube metadata/captions
4. Browser MCP — JS-heavy platforms (Vimeo showcases, Substack) when WebFetch fails

Enumerate to exhaustion: channel RSS, podcast feed, blog sitemap, author-archive pagination — not page one.

Key rules

1. Everything except books. Comprehensive across all 14 player categories in players-checklist.md.
2. Observe before searching. Read docs/movement_leader_research/ dossier first; never re-fetch docs/articles/, docs/videos/, docs/harvest/.
3. "Checked, none found" is required — record in tracker and sources-swept.md.
4. De-duplicate across platforms. One canonical item + alternate URLs.
5. Migration, not creation. FETCH files are verbatim source with attribution front-matter. No fabricated URLs, dates, or titles.
6. Persist as you go. Write rows and files before ending a step or turn.
7. No app code changes. Only write under docs/movement_leader_research/content/, docs/harvest/, and existing docs/articles|videos/ conventions.
8. Gate honestly. Mark verified in step 09 only when all 8 coverage checks pass; update CONTENT_REGISTRY.md.

Session report (print when done)

When the pipeline finishes (or stops at --to-step), print:

1. TENANT_SLUG and final tracker status (verified / harvested / swept / blocked)
2. Categories swept: N/14
3. Items by disposition: FETCH / LINK / TRANSCRIBE / SKIP
4. Files written to docs/harvest/ (count + find verification)
5. Transcribe queue size and any gate failures
6. CONTENT_REGISTRY.md row updated (if step 09 passed)

Additional resources

• 14 player categories: reference/players-checklist.md
• Pipeline diagram + batch strategy: docs/build/skills/content-gathering/MASTER_PLAYBOOK.md
• Cross-tenant status: docs/build/skills/content-gathering/CONTENT_REGISTRY.md
• Tracker template: docs/build/skills/content-gathering/HARVEST_TRACKER_TEMPLATE.md

Downstream pipeline (after step 09)

Harvest ends at repo inventory. For Supabase upload:

Books remain movemental-leader-corpus-upload, not this pipeline.