Routine — Investor Credibility Ingest

A subroutine of crawl-fetch-ingest. Where the parent skill’s primary walk is firm-anchored (one VC firm → its team → its portfolio → those portcos’ CEOs), this routine inverts the starting point: company-anchored (one operating company → its list of backer firms → each backer’s team + portfolio).

Both walks are first-class. The skill’s cascade (Jina / Firecrawl / OpenGraph.io / Brandfetch / SVG-tier fallback / bg-strip) is shared. Only the entry point and stop condition differ.

The perspective

The deck a reader is about to look at will name several VC firms most readers have never heard of. The skill’s job here is to make those firms — and by extension the round itself — legible, by surfacing two cheap-to-fetch signals per firm:

1. Who’s on the team (CP1) — a few partner names + photos. “These are real people, not a brand.”
2. Who they’ve already backed (CP3) — a row of recognizable portfolio logos. “If those companies trust this firm, this round is in good company.”

We deliberately stop at the portfolio companies — we do not descend to portco CEOs (CP4 in the parent skill). At credibility-card distance, portco-CEO faces don’t add legibility; logos do.

When to invoke

User says something like:

• “ingest our backers”
• “fill out the investor section”
• “do credibility ingest on these firms”
• “we need to make these VCs legible to readers who don’t know them”
• “go get teams and portfolios for {Firm A}, {Firm B}, {Firm C}”
• “{operating company} is raising — give us context on the investors”

The user typically supplies a list of firm names (Quiet Capital, Bloomberg Beta, Air Street Capital, AIX Ventures) along with the context company they all back (e.g., Chroma). The context company is metadata — we don’t crawl it here; the parent skill’s team/ flow already handles the operating company’s own people.

Inputs

• Required: list of backer firms. Names or URLs. If only names, resolve homepage via Firecrawl-search before proceeding.
• Optional: context company slug — gets recorded as ingested_for: <company-slug> in each firm’s firm.md so future readers know why this data exists.
• Optional: --marquee-portcos N — cap portfolio capture at N companies per firm (default: capture all the firm lists; the rendering layer chooses which subset to display). Useful when a firm has 200+ portfolio companies and you only need ~12 marquee logos.

Output layout

<cwd>/data/investors/{firm-slug}/
  firm.md                            # firm-level metadata; includes ingested_for
  team/{person-slug}.md              # CP1: every partner / venture partner / op partner / EIR / advisor
  team/{person-slug}.{jpg|png}       # headshot
  portfolio/{co-slug}.md             # CP3: every portfolio company surfaced
  portfolio/trademark__{Co-Name}.svg # primary logo (SVG preferred; raster fallback per cascade)
  portfolio/favicon__{Co-Name}.{png|svg}
  # NO {co-slug}-ceo.md — CP4 is skipped for this routine

The data/investors/ directory parallels data/firms/ (the firm-anchored anchor) and data/team/ (operating-company team from the inverse run). All three can coexist in the same project — a fundraise deck typically has all three populated:

<project>/data/
  team/                       # the operating company's own employees (company-anchored run, CP1-only)
  investors/                  # backer firms (this routine)
    {firm-a}/firm.md + team/ + portfolio/
    {firm-b}/firm.md + team/ + portfolio/
  firms/                      # (only if the project also ran the firm-anchored walk on some specific VC)
    {firm}/firm.md + team/ + portfolio/ + portco-ceos/

If the project ALSO has a top-level data/team/ for the operating company, that stays untouched — data/investors/{firm}/team/ is namespaced under each backer firm.

Workflow for Claude

For each firm in the input list, run the parent skill’s checkpoints with the following scope:

Per firm

1. Discover firm site — Firecrawl-search if only a name was given; confirm homepage with the user if ambiguous.
2. Write firm.md with ingested_for: <company-slug> and ingest_routine: investor-credibility-ingest.
3. CP1 (team) — enumerate every role-bearing sub-page per the parent skill’s page-discovery cascade (/team, /partners, /venture-partners, /operating-partners, /eir, /supporting-partners, /lpac, /advisors, /about/team, paginated variants). Capture all partner-level people; the role_class field distinguishes managing-partner / venture-partner / operating-partner / EIR / supporting-partner / advisor. Headshots via OpenGraph.io on the per-person bio URL; LinkedIn via Firecrawl with stealth proxy.
4. CP3 (portfolio) — enumerate /portfolio, /companies, /investments. For each company:
   • Trademark via the 7-tier cascade (inline SVG → site SVG paths → press/brand kit → Brandfetch → SVG-repo Tavily → CSE → raster + bg-strip).
   • Favicon via the favicon cascade (apple-touch-icon → site icons → S2 fallback).
   • og:image URL captured into frontmatter (no file save).
   • Company .md with sector, stage, description, website.
5. Skip CP4.
6. Cache aggressively. Same ~/.claude/skills/crawl-fetch-ingest/cache/{firm-slug}/ convention as the parent skill — re-running this routine on the same firm is free.

Across firms

• Run firms in sequence, not parallel at the routine level (each firm’s CP1 + CP3 can parallelize internally). One firm’s batch is a natural checkpoint; pause between firms only if the user is reviewing in real-time.
• De-duplicate portfolio overlap. Two firms can co-invest in the same company. If data/investors/{firm-a}/portfolio/notion.md and data/investors/{firm-b}/portfolio/notion.md both exist, that’s fine and intentional — each firm’s portfolio is its own narrative. Don’t try to canonicalize.
• Marquee selection is a rendering concern, not an ingest concern. Capture the full portfolio; the deck/site picks the 6–12 logos to display. Don’t pre-filter at ingest.

Confidence + flagging

Same rubric as the parent skill (high / medium / low / flagged). For credibility-card use, the bar to publish is higher than for an internal one-pager — partner photos that look like generic LinkedIn placeholders, or logos that bg-strip mangled, should land as low or flagged so the human reviewer catches them.

Reporting

When done, return a summary covering:

• Per-firm: team count + portfolio count + asset-success rate (logos SVG vs. raster vs. fallback).
• Cross-cutting flags — portfolio companies whose logo only resolved as raster with bg-strip (will likely look uneven next to clean SVGs in a row), partners whose headshot couldn’t be resolved (will leave a gap in a partner-grid render).
• Notable overlaps — if multiple backer firms co-invest in the same well-known company, surface it; that’s a credibility multiplier the rendering layer may want to highlight.

Anti-patterns

• Don’t crawl portfolio-company team pages. That’s CP4 of the firm-anchored walk; deliberately out of scope here. A credibility card showing “Notion’s CEO is X” is too deep — readers care that this firm backed Notion, not who runs Notion.
• Don’t paraphrase from training data when the firm has a public site. Cache says no? Fetch the site. Always.
• Don’t try to rank firms by quality. This routine surfaces facts (team + portfolio); ranking is editorial and belongs in the deck-authoring step, not the ingest.
• Don’t merge investor data into the operating company’s own data/team/. Backers live in data/investors/{firm}/team/; the company’s own employees live in data/team/. Keeping them separate makes both renderings cleaner.

See also

• ../SKILL.md — the parent skill, the cascade definitions, the two anchor types
• ../schema/firm.md — firm-level frontmatter
• ../schema/person.md — person frontmatter incl. role_class taxonomy
• ../schema/company.md — portfolio-company frontmatter incl. logo asset_strategy
• ./triage-brand-assets.md — quality-review subroutine to run on data/investors/{firm}/portfolio/ once this routine finishes (often the immediately next step)