From cfa-core
Manages CFA managed-agent cookbooks for Anthropic API: lists by cost tier, validates schemas, audits skill coverage, assembles deploy payloads. Rust tooling docs.
How this skill is triggered — by the user, by Claude, or both
Slash command
/cfa-core:cfa-managed-agentThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
You have access to the managed-agent cookbook tooling for CFA agents. **All work happens in deterministic Rust** in `crates/corp-finance-core/src/managed_agent/`. This skill describes the available surfaces — CLI, MCP tools, NAPI bindings — and when to use which. The skill itself contains no business rules; if you need to know what makes a cookbook valid, read the Rust source or call the valida...
You have access to the managed-agent cookbook tooling for CFA agents. All work happens in deterministic Rust in crates/corp-finance-core/src/managed_agent/. This skill describes the available surfaces — CLI, MCP tools, NAPI bindings — and when to use which. The skill itself contains no business rules; if you need to know what makes a cookbook valid, read the Rust source or call the validate tool.
CFA cookbooks are tagged by cost tier so users can run the system without paying for any vendor data subscription:
| Tier | Meaning |
|---|---|
core_only | Runs against cfa-core MCP only. User supplies inputs as JSON. Always free. |
freemium | cfa-core + free public data (FRED, EDGAR, FIGI, YF, WB, geopolitical) and/or FMP free tier. No paid subscription required. |
paid_vendor | Requires LSEG, S&P Global, FactSet, Morningstar, Moody's, PitchBook, Aiera, or Daloopa subscription. User must supply credentials at deploy time. |
To enumerate cookbooks by tier, prefer the MCP tool over scraping the registry:
managed_agent_list({ tier: "core_only" }) # 5 cookbooks
managed_agent_list({ tier: "freemium" }) # 8 cookbooks
managed_agent_list({ tier: "paid_vendor" }) # 2 cookbooks
managed_agent_list({}) # all 15, grouped
For users on the free tier, never suggest a paid_vendor cookbook unless they've explicitly indicated they have the vendor credentials. The 13 non-paid cookbooks cover equity research, IB pitch / merger / LBO, PE deal screening / IC memo / LP audit, wealth planning, KYC, GL recon, model audit, and earnings analysis end-to-end.
All exposed by the cfa-core MCP server:
| Tool | What it does |
|---|---|
managed_agent_list | Enumerate cookbooks by cost tier. Registry-only, no I/O. |
managed_agent_validate | Schema-validate one cookbook. Checks slug allowlist, agent.json shape, system prompt + skill + subagent path resolution, steering examples. Returns per-check pass/fail. |
managed_agent_check_all | Walk every cookbook under cookbooks_root and validate each. Returns aggregate counts + per-cookbook outcome. |
managed_agent_sync | Coverage audit between cookbooks and skills. Flags missing skills (referenced but not on disk) and orphan skills (on disk but unreferenced). Does NOT copy files — skills are resolved at deploy time in our pattern. |
managed_agent_deploy | Assemble a deploy-ready JSON payload for one cookbook: ${VAR} substitution from env_vars, system-prompt inlining, skill resolution. Returns orchestrator + subagent + skill payloads. The consumer POSTs to /v1/agents. |
managed_agent_orchestrate | Route a single handoff_request event to a target cookbook. Validates event type and target slug against the allowlist. |
cfa managed-agent list # all cookbooks
cfa managed-agent list --tier=core-only # zero-dependency cookbooks
cfa managed-agent list --tier=freemium
cfa managed-agent list --tier=paid-vendor
cfa managed-agent validate <slug> # one cookbook
cfa managed-agent check-all # every cookbook
cfa managed-agent sync # cookbook ↔ skill coverage
cfa managed-agent deploy <slug> # assembled JSON to stdout
cfa managed-agent orchestrate <slug> # stdin event-loop router
scripts/test-cookbooks.sh runs validate + dry-run deploy on every cookbook (CI smoke harness).
managed_agent_list MCP tool with the appropriate tier filter.managed_agent_validate MCP tool, or CLI cfa managed-agent validate <slug>.scripts/test-cookbooks.sh.managedAgentDeploy(JSON.stringify(input)) via the NAPI binding directly (no MCP overhead).There is one and only one source of truth for what a valid cookbook is: crates/corp-finance-core/src/managed_agent/validate.rs. The CLI, MCP tool, and NAPI binding are pure adapters. Do not duplicate validation rules in this skill, in cookbook READMEs, or anywhere else — call the validator.
managed-agent-cookbooks/<slug>/agent.json + subagents/*.json + steering-examples.json + README.mdfrom_skill: <slug> resolves to .claude/skills/<slug>/SKILL.mdsystem.file resolves first to cookbook-relative, then to .claude/agents/cfa/<name>.mdcrates/corp-finance-core/src/managed_agent/types.rs (COOKBOOK_REGISTRY + ALLOWED_SLUGS)The upstream Python tooling shipped validate-cookbook.py and check.py as separate scripts. Both fold into one Rust function (validate::validate_manifest) and one batched walker (validate::validate_all). If you find yourself wanting to add a new validation rule, add it to validate.rs and write a Rust unit test, not a markdown checklist.
Provides UI/UX resources: 50+ styles, color palettes, font pairings, guidelines, charts for web/mobile across React, Next.js, Vue, Svelte, Tailwind, React Native, Flutter. Aids planning, building, reviewing interfaces.
Fetches up-to-date documentation from Context7 for libraries and frameworks like React, Next.js, Prisma. Use for setup questions, API references, and code examples.
npx claudepluginhub fall-development-rob/corp_finance --plugin cfa-core