Verdict. The repo layout is the part to copy. Its sync script keeps skills from drifting across agents, which is the piece most agent codebases are missing. The agents themselves are only useful with paid data seats behind them.

what it is

Anthropic’s official open-source reference stack for financial-services work. It has 10 named agents, 7 vertical skill bundles, 11 MCP data connectors and Managed Agent cookbooks. You can install it as Cowork or Claude Code plugins, or deploy it headless through /v1/agents with the same system prompt and the same skills.

The layout has three layers.

  1. Agents (plugins/agent-plugins/<slug>/). Self-contained workflow plugins, each bundling its own skills. This is what the end user sees.
  2. Vertical plugins (plugins/vertical-plugins/<vertical>/). The source of truth for skills, slash commands and .mcp.json connectors, written once and synced into the agents.
  3. Managed-agent cookbooks (managed-agent-cookbooks/<slug>/). An agent.yaml, depth-1 leaf subagents and steering examples. Same prompt, headless harness.

scripts/sync-agent-skills.py pushes skill edits from the verticals into the agents. scripts/check.py validates cross-references before a push.

VerticalAgents
Coverage & advisoryPitch Agent, Meeting Prep Agent
Research & modelingMarket Researcher, Earnings Reviewer, Model Builder
Fund admin & finance opsValuation Reviewer, GL Reconciler, Month-End Closer, Statement Auditor
Operations & onboardingKYC Screener

Each agent ships with a Managed Agent counterpart, deployed with scripts/deploy-managed-agent.sh <slug>.

The skill bundles:

  • financial-analysis (core, includes .mcp.json). /comps, /dcf, /lbo, /3-statement-model, /debug-model, pptx-author, xlsx-author, ppt-template-creator, skill-creator
  • investment-banking. /one-pager, /cim, /teaser, /buyer-list, /merger-model, /process-letter, /deal-tracker
  • equity-research. /earnings, /earnings-preview, /initiate, /model-update, /morning-note, /sector, /thesis, /catalysts, /screen
  • private-equity. /source, /screen-deal, /dd-checklist, /dd-prep, /unit-economics, /returns, /ic-memo, /portfolio, /value-creation, /ai-readiness
  • wealth-management. /client-review, /financial-plan, /rebalance, /client-report, /proposal, /tlh
  • fund-admin. Reconciliations, accruals, roll-forwards, NAV tie-out
  • operations. KYC document parsing, rules-grid evaluation

The 11 hosted MCP connectors are all subscription-gated. They are Daloopa, Morningstar, S&P Global (Kensho and Capital IQ), FactSet, Moody’s, MT Newswires, Aiera, LSEG, PitchBook, Chronograph and Egnyte. LSEG and S&P Global also contribute partner-built plugins under plugins/partner-built/.

Excel and PowerPoint work two ways. An admin plugin (claude-for-msft-365-install/) sets up the live M365 add-in, with the Anthropic API, Vertex, Bedrock or an internal gateway as the backend. Or the pptx-author and xlsx-author skills generate .pptx and .xlsx files headless inside Managed Agents, with no Office runtime.

how I use it

I have three of the bundles enabled as Claude Code plugins (financial-analysis, private-equity and investment-banking).

what I found

  • The private-equity bundle covers a lot of deal workflow out of the box, from /source and /screen-deal through /ic-memo and /value-creation. A deal screener on its own is no longer a differentiator.
  • The connector list maps the Bloomberg-replacement seats Anthropic is courting. There’s no Tegus, no AlphaSense and no CRE data (CoStar, Reonomy, Placer).
  • The M365 install plugin targets the IT admin, not the analyst. That’s a distribution route into the enterprise.
  • Porting the sync, check and orchestration scripts to another stack is real engineering. I’d estimate about two weeks.
  • Adopting the skill format ties you to Anthropic’s plugin and Managed Agent runtime. Keeping the skill markdown portable is the hedge.