company-manifest-queue

name	company-manifest-queue
description	Use when the user wants to draft or queue new company manifests for this repo, either for one company or a confirmed batch from an index, sector, region, or arbitrary grouping. It maps draft company data onto the repo's CompanyManifest schema and stages queue entries with `bun run company:queue` instead of writing canonical manifests directly.

Company Manifest Queue

Use this skill for net-new company intake. The goal is to create valid CompanyManifest drafts and stage them in content/manifests/queue/, not to publish canonical manifests directly.

If the user only gives raw company names, tickers, or group phrases, route that request through bun run company:intake --mode=prepare instead of requiring a full manifest draft up front.

Source of Truth

Before drafting anything, read these repo files:

• src/lib/domain/content-types.ts for the CompanyManifest contract
• content/taxonomy/regions.json
• content/taxonomy/indices.json
• content/taxonomy/sectors.json
• content/taxonomy/industries.json
• content/taxonomy/technology-waves.json
• 2-3 examples from content/manifests/companies/*.json
• scripts/company-queue.ts and scripts/company-init.ts for the intake flow

Do not invent taxonomy IDs. If the repo taxonomy does not support the requested company cleanly, surface that gap instead of guessing.

Modes

Single-company intake

Use when the user names one company or ticker.

Collect or confirm:

• company name and ticker
• slug
• region, index, sector, industry
• CompaniesMarketCap URL
• short company description
• relevant technology wave IDs
• optional IPO metadata
• optional seed product names
• optional seed source URLs
• notes that explain why this company is entering the queue

Batch intake

Use when the user wants a set of companies from an index, sector, region, theme, or arbitrary grouping.

Workflow:

1. If the user provided only raw names, tickers, or a group phrase, start with bun run company:intake --mode=prepare.
2. Confirm or narrow the list with the user if the grouping is ambiguous and they want manual control.
3. Draft one manifest per confirmed company.
4. Stage them with a consistent --batch-id and --group-label.

Do not queue ambiguous candidates without confirmation.

Queueing Rules

• Check content/manifests/companies/ and content/manifests/queue/ first.
• If a slug is already canonical or already queued, stop and tell the user unless they explicitly asked to replace it.
• Write plain CompanyManifest draft JSON first.
• Stage drafts with bun run company:queue --manifest=....
• For raw intake requests, let company:intake create content/manifests/unverified/<request-id>.json, resolve candidates, and queue only the manifests that validate cleanly.
• When the raw intake overlaps companies that already have published research, ask whether the user wants to skip them or refresh them. Pass that choice with --already-researched=skip|refresh.
• In batch mode, include --batch-id, --group-label, and --request-notes when useful.
• If the user wants the whole intake flow handled end-to-end, prefer bun run company:pipeline --manifest=... --provider=auto instead of stopping after queueing.
• If the user wants the raw-intake flow handled end-to-end, prefer bun run company:intake --raw=... --mode=dry-run --provider=auto or --mode=publish when they explicitly asked for publication.
• Otherwise end by telling the user the next command: bun run company:init --queued=<slug>.
• Mention the follow-up Ralph step after promotion: bun run sync:company --company=<slug> --mode=dry-run.

After a company:intake --mode=prepare run, the normal inline follow-up is:

• already researched: skip or refresh
• queue/prepared only
• research dry-run
• publish website

If the user already requested dry-run or publish in the original prompt, skip the later workflow prompt and honor the requested end state directly. If they did not specify what to do with already researched companies, ask that first.

Guardrails

• Never write directly to content/manifests/companies/ from this skill.
• Prefer factual notes over narrative prose.
• Keep seedProductNames and seedSourceUrls as hints, not exhaustive research.
• If the taxonomy is missing the needed sector, industry, index, region, or wave, stop and surface the exact missing IDs instead of creating ad hoc values.