// the find
anthropics/financial-services
Anthropic's official collection of Claude plugins and agent templates for financial services workflows — investment banking, equity research, PE, fund admin, and wealth management. It targets financial analysts and IT teams at financial firms who want to wire Claude into their existing tooling (Excel, PowerPoint, data terminals) rather than build from scratch. Everything is markdown and YAML prompt templates, no application code.
- The dual-deployment model (Cowork plugin vs. Managed Agents API from one source) is genuinely useful — firms can start with a no-code install and graduate to headless API deployment without rewriting anything.
- MCP connector list is solid and current: FactSet, Bloomberg-adjacent (LSEG), PitchBook, Morningstar, Moody's — the actual data sources practitioners use, not placeholder examples.
- The check.py/validate.py/sync-agent-skills.py toolchain prevents the bundled skill copies from drifting out of sync with vertical sources, which is the exact maintenance problem that kills repos like this long-term.
- Compliance posture is appropriately conservative — the disclaimer is prominent and specific about what the agents explicitly do not do (execute trades, bind risk, post to ledger), which is important given the regulatory environment.
- The entire thing is prompt templates and YAML — there is no testable logic, no unit tests for the skill files, and no way to verify the DCF or LBO outputs are numerically correct before putting them in front of a client. The validate_dcf.py script exists but appears to be a structural linter, not a financial correctness check.
- KYC Screener is the highest-stakes agent here and gets the thinnest treatment — rules-engine.yaml is a prompt file, not an actual rules engine. Firms with real AML/KYC obligations will need to replace the core logic entirely, which the README undersells.
- Heavy dependency on Claude-specific infrastructure (Cowork, Managed Agents API, Claude Code plugin system) that doesn't exist yet in public docs as of early 2025 — adopters outside the Anthropic ecosystem have no migration path and are fully locked in.
- Subagent delegation is explicitly labeled a 'Research Preview' but is used in the most complex agents (Pitch Agent, GL Reconciler) without clear fallback behavior documented if the capability changes or breaks.