# Claude Finance — The Shape of a Skill

> Anthropic shipped a skills library *for finance*. Read it slowly.

A structural reading of [`anthropics/financial-services`](https://github.com/anthropics/financial-services) — 213 skill.md files released May 2026, Apache 2.0. This page maps every skill to one of eighteen architectural archetypes (Foundation + 17 active) grouped into four families of work.

Part of the [Cartu collection](https://cartu.app) — translation experiments that draw the shape of dense published artifacts.

Original analysis: T Ngo / HelloTNgo · Live at [claudefinance.cartu.app](https://claudefinance.cartu.app)

---

## The Frame

Most published work on AI agent skills is by builders, for builders. Heavy on syntax. Light on visual mental models.

*The Shape of a Skill* is an attempt to lower the barrier. The reader knows the shape of the territory before any single skill enters the picture.

This is a translation experiment. Use it as a map. Cartu is the method behind it.

Because trust and control sit downstream of comprehension.

The *"I don't know"* reaction to a new AI artifact is the barrier underneath every other adoption barrier. Naming the shape of the skill lowers the barrier to adoption and increases comprehension.

---

## Layer 0 — The Whole Library

The wide-angle view. 213 skill.md files mapped to the eighteen archetypes.

### Distribution by family

| Family | Count | Members |
|--------|-------|---------|
| **Foundation** | 59 | the shared library — see Layer 1 |
| **Navigators** | 68 | Front desk · Orchestrator · Gatekeeper · Vendor pipeline · Setup |
| **Gatherers** | 12 | Researcher · Monitor · Ledger · Catalog |
| **Analyzers** | 35 | Modeler · Reconciler · Rater · Triager · Diligence |
| **Makers** | 39 | Drafter · Redliner · Production |
| **Total** | **213** | |

### Top archetypes by file count

| Archetype | Count | Notes |
|-----------|-------|-------|
| Foundation | 59 | the biggest single archetype — the layer everything else plugs into |
| Front desk | 46 | thin slash-command wrappers across verticals (22% of the library) |
| Drafter | 25 | narrative producers — memos, IM, client reports |
| Modeler | 14 | DCF, LBO, 3-statement, merger, returns |
| Diligence | 13 | parallel-framework investigation with explicit gaps |
| Orchestrator | 11 | one per agent plugin — routes to sub-skills |
| Production | 9 | slot-fillers for xlsx / pptx / docx |
| Reconciler | 8 | match-and-classify, root-cause, recompute, foot-check |
| Vendor pipeline | 8 | LSEG MCP tool chains with directional output |
| Researcher | 5 | external data gathering with citation discipline |
| Redliner | 5 | guarded section edits |
| Triager | 5 | bucket-classifier with playbook (e.g., pass/diligence/no) |
| Setup | 3 | scaffolds the user's own work (skill-creator, dd-meeting-prep) |
| Monitor | 3 | event-driven state sweeps |
| Ledger | 3 | state across init/update/review/close |
| Rater | 2 | rule-citation scoring (KYC) |
| Catalog | 1 | structured extraction (kyc-doc-parse) |
| Gatekeeper | 0 primary | appears as secondary only (initiating-coverage) |

### The Lookup — every skill by domain

### Agent Plugins · 70 skills · 10 agents · each agent has an orchestrator + sub-skills


**earnings reviewer** (7 skills)

- `audit-xls` — **Foundation** — scope-tiered spreadsheet audit; reports first, fixes on request
- `earnings analysis` — **Drafter** — 8-12 page DOCX equity update; date-verification; refuses distribution
- `earnings-preview` — **Drafter** — 1-page pre-earnings preview with bull/base/bear + catalyst checklist
- `earnings-reviewer` — **Orchestrator** — routes to 5 sub-skills; treats transcripts as untrusted; never publishes
- `model-update` — **Redliner** — plug new actuals into existing model; old-vs-new delta; reconcile first
- `morning-note` — **Drafter** — 1-page time-stamped morning meeting note; no view = useless
- `xlsx-author` — **Foundation** — headless .xlsx file emitter (shared sub-skill)


**gl reconciler** (5 skills)

- `audit-xls` — **Foundation** — shared spreadsheet audit
- `break-trace` — **Reconciler** — root-cause a single break; JSON with owner + clear-date; diagnostic-only
- `gl-recon` — **Reconciler** — full-outer-join match into 6 buckets with tolerance; hypothesis only
- `gl-reconciler` — **Orchestrator** — per-asset-class reader dispatch; orchestrator never writes
- `xlsx-author` — **Foundation** — shared headless authoring


**kyc screener** (4 skills)

- `kyc-doc-parse` — **Catalog** + **Triager** — structured JSON extraction from onboarding packet; null for missing
- `kyc-rules` — **Rater** — apply firm's rules grid; per-rule outcome with citation; routes for approval
- `kyc-screener` — **Orchestrator** — doc-reader has Read/Grep only; length-capped JSON; never approves
- `xlsx-author` — **Foundation** — shared headless authoring


**market researcher** (6 skills)

- `competitive-analysis` — **Drafter** — 2-phase competitive deck; refuses slides until outline approved
- `comps-analysis` — **Production** — institutional-grade comps spreadsheet; MCP-data-source hierarchy
- `idea-generation` — **Triager** — 5 named screen playbooks; screens surface candidates, not conclusions
- `market-researcher` — **Orchestrator** — multi-step pipeline with 2 stop-and-surface checkpoints
- `pptx-author` — **Foundation** — headless .pptx emitter (shared sub-skill)
- `sector-overview` — **Researcher** + **Drafter** — industry landscape report with TAM/competitive/thematic sections


**meeting prep agent** (5 skills)

- `client-report` — **Drafter** + **Production** — 9-section performance report; compliance gate
- `client-review` — **Drafter** — one-page summary + agenda + action items
- `investment-proposal` — **Drafter** + **Production** — 6-section prospect deck; personalized, not templated
- `meeting-prep-agent` — **Orchestrator** — calendar-event-bound; news-reader worker has no MCP access
- `pptx-author` — **Foundation** — shared headless authoring


**model builder** (7 skills)

- `3-statement-model` — **Modeler** — IS/BS/CF template population; 6 verification checkpoints; BS balance
- `audit-xls` — **Foundation** — shared audit
- `comps-analysis` — **Production** — shared comps
- `dcf-model` — **Modeler** — WACC, terminal value, 75-cell sensitivity; recalc.py before delivery
- `lbo-model` — **Modeler** — template-required; 5-section gated workflow; 4-color font convention
- `model-builder` — **Orchestrator** — routes to 4 model-type sub-skills; every output is a formula
- `xlsx-author` — **Foundation** — shared headless authoring


**month end closer** (6 skills)

- `accrual-schedule` — **Modeler** + **Drafter** — per-accrual derivation table + draft JE; refuses to post
- `audit-xls` — **Foundation** — shared audit
- `month-end-closer` — **Orchestrator** — per-schedule worker dispatch; no GL posting
- `roll-forward` — **Reconciler** — 7-line roll-forward template; must foot or surface unexplained item
- `variance-commentary` — **Drafter** + **Triager** — flux commentary for material lines; flag unclear drivers
- `xlsx-author` — **Foundation** — shared headless authoring


**pitch agent** (21 skills)

- `3-statement-model` — **Production** — explicit template-completion mode (slot-filler)
- `3-statement-model_reference_formatting` — **Foundation** — reference asset
- `3-statement-model_reference_formulas` — **Foundation** — reference asset
- `3-statement-model_reference_sec-filings` — **Foundation** — reference asset
- `audit-xls` — **Foundation** — shared audit
- `comps-analysis` — **Production** — shared comps
- `dcf-model` — **Modeler** — with correct_patterns / common_mistakes blocks
- `deck-refresh` — **Redliner** — 4-phase number swap with Phase 3 approval gate
- `investment bank check deck` — **Diligence** — 4-dimension parallel sweep; severity-graded; refuses to fix
- `investment bank check deck_reference_ib-report format` — **Foundation** — reference asset; report template
- `investment bank check deck_reference_ib-terminology` — **Foundation** — reference asset; casual->professional replacements
- `lbo-model` — **Production** — slot-filler mode (template-completion)
- `pitch deck` — **Production** — template-first; 5-phase progress with LibreOffice validate-fix loop
- `pitch deck_reference_calculation-standards` — **Foundation** — reference asset
- `pitch deck_reference_formatting-standards` — **Foundation** — reference asset
- `pitch deck_reference_slide-templates` — **Foundation** — reference asset; data-template mismatch playbook
- `pitch deck_reference_xml-reference` — **Foundation** — reference asset; OOXML patterns + anti-XML guardrails
- `pitch-agent` — **Orchestrator** — top-level routing; checkpoints after Excel + after deck
- `pptx-author` — **Foundation** — headless emitter (shared sub-skill)
- `sector-overview` — **Researcher** — TAM with hype-vs-realistic; staleness flagging
- `xlsx-author` — **Foundation** — headless emitter (shared sub-skill)


**statement auditor** (4 skills)

- `audit-xls` — **Foundation** — shared audit
- `nav-tieout` — **Reconciler** — independent recompute of LP capital account; pass/fail per line
- `statement-auditor` — **Orchestrator** — sub-agent has Read/Grep only; pass/hold; refuses to send
- `xlsx-author` — **Foundation** — headless emitter (shared sub-skill)


**valuation reviewer** (5 skills)

- `ic-memo` — **Drafter** — 9-section memo with page budgets; ask for missing inputs
- `portfolio-monitoring` — **Triager** + **Monitor** — GREEN/YELLOW/RED variance triage + covenant status
- `returns-analysis` — **Modeler** — 2-way IRR/MOIC sensitivities; Bull/Base/Bear; firm tolerances
- `valuation-reviewer` — **Orchestrator** — untrusted-input boundary; no external distribution
- `xlsx-author` — **Foundation** — headless emitter (shared sub-skill)


### Partner Built · 25 skills · LSEG (16) + S&P Global (9) · vendor integrations


**lseg** (16 skills)

- `bond-futures-basis` — **Vendor pipeline** — prices bond future, CTD, basis/repo; long/short recommendation
- `bond-relative-value` — **Vendor pipeline** — spread decomposition; rich/cheap call
- `commands_analyze-bond-basis` — **Front desk** — slash-command wrapper; delegates to bond-futures-basis
- `commands_analyze-bond-rv` — **Front desk** — slash-command wrapper; delegates to bond-relative-value
- `commands_analyze-fx-carry` — **Front desk** — slash-command wrapper; delegates to fx-carry-trade
- `commands_analyze-option-vol` — **Front desk** — slash-command wrapper; delegates to option-vol-analysis
- `commands_analyze-swap-curve` — **Front desk** — slash-command wrapper; delegates to swap-curve-strategy
- `commands_macro-rates` — **Front desk** — slash-command wrapper; delegates to macro-rates-monitor
- `commands_research-equity` — **Front desk** — slash-command wrapper; delegates to equity-research
- `commands_review-fi-portfolio` — **Front desk** — slash-command wrapper; delegates to fixed-income-portfolio
- `equity-research` — **Vendor pipeline** — IBES consensus + fundamentals + macro -> buy/hold/sell
- `fixed-income-portfolio` — **Vendor pipeline** — prices bond basket; scenarios; portfolio review
- `fx-carry-trade` — **Vendor pipeline** — spot + forwards + vol -> carry-to-vol + tenor recommendation
- `macro-rates-monitor` — **Vendor pipeline** — macro + curves + breakevens -> rates dashboard
- `option-vol-analysis` — **Vendor pipeline** — vol surface + realized vol -> IV-RV premium
- `swap-curve-strategy` — **Vendor pipeline** — curve build + metrics -> DV01-neutral trade recommendation


**spglobal** (9 skills)

- `earnings-preview-beta` — **Researcher** + **Drafter** — 8-phase research-to-HTML; hyperlink-every-claim; allowlist
- `earnings-preview-beta_reference_report-template` — **Foundation** — reference asset; frozen HTML+CSS+JS template
- `funding-digest` — **Researcher** + **Production** — universe-building + PPTX slide; 6 entity-resolution rules
- `funding-digest_reference_sector-seeds` — **Foundation** — reference asset; vendor dictionary
- `tear-sheet` — **Orchestrator** + **Production** — audience-routed DOCX generator (ER / IB-M&A / Corp Dev / Sales-BD)
- `tear-sheet_reference_corp-dev` — **Foundation** — reference asset; audience plan
- `tear-sheet_reference_equity-research` — **Foundation** — reference asset; audience plan
- `tear-sheet_reference_investment bank-ma` — **Foundation** — reference asset; audience plan
- `tear-sheet_reference_sales-broker dealer` — **Foundation** — reference asset; audience plan


### Vertical Plugins · 118 skills · 7 verticals · domain-specific shapes per practice area


**equity research** (29 skills)

- `catalyst-calendar` — **Monitor** — forward-looking event schedule across coverage universe
- `commands_catalysts` — **Front desk** — slash-command wrapper; delegates to catalyst-calendar
- `commands_earning` — **Drafter** — outlier: full Drafter inlined into the command file
- `commands_earnings-preview` — **Front desk** — slash-command wrapper; delegates to earnings-preview
- `commands_initiate` — **Front desk** — slash-command wrapper; delegates to initiating-coverage
- `commands_model update` — **Front desk** — slash-command wrapper; delegates to model-update
- `commands_morning note` — **Front desk** — slash-command wrapper; delegates to morning-note
- `commands_screen` — **Front desk** — slash-command wrapper; delegates to idea-generation
- `commands_sector` — **Front desk** — slash-command wrapper; delegates to sector-overview
- `commands_thesis` — **Front desk** — slash-command wrapper; delegates to thesis-tracker
- `earnings-analysis` — **Drafter** + **Researcher** — 8-12 page DOCX; date-freshness gate
- `earnings-analysis_reference_best-practices` — **Foundation** — reference asset
- `earnings-analysis_reference_report-structure` — **Foundation** — reference asset; template
- `earnings-analysis_reference_workflow` — **Foundation** — reference asset; procedure
- `earnings-preview` — **Drafter** — 1-page preview with bull/base/bear
- `idea-generation` — **Triager** — 5 named screens; ranked shortlist
- `initiating-coverage` — **Ledger** + **Gatekeeper** — 5-task workflow with one-task-per-request gating
- `initiating-coverage_asset_quality-checklist` — **Foundation** — reference asset; QC asset
- `initiating-coverage_asset_report-template` — **Foundation** — reference asset; institutional template
- `initiating-coverage_reference_task1-company-research` — **Foundation** — reference asset; Researcher-shaped procedure
- `initiating-coverage_reference_task2-financial-modeling` — **Foundation** — reference asset; Modeler-shaped procedure
- `initiating-coverage_reference_task3-valuation` — **Foundation** — reference asset; Modeler-shaped procedure
- `initiating-coverage_reference_task4-chart-generation` — **Foundation** — reference asset; Production-shaped procedure
- `initiating-coverage_reference_task5-report-assembly` — **Foundation** — reference asset; Drafter + Production procedure
- `initiating-coverage_reference_valuation-methodologies` — **Foundation** — reference asset; pedagogical reference
- `model-update` — **Redliner** — guarded section-level model edits
- `morning-note` — **Monitor** + **Drafter** — daily overnight sweep; opinion required
- `sector-overview` — **Drafter** + **Researcher** — industry landscape
- `thesis-tracker` — **Ledger** — init/update/review/check modes around investment thesis


**financial analysis** (29 skills)

- `3-statement-model` — **Modeler** — IS/BS/CF template; verification checkpoints; BS balance
- `3-statement-model_reference_formatting` — **Foundation** — reference asset
- `3-statement-model_reference_formulas` — **Foundation** — reference asset
- `3-statement-model_reference_sec-filings` — **Foundation** — reference asset
- `audit-xls` — **Foundation** — scope-tiered spreadsheet audit
- `clean-data-xls` — **Redliner** — proposal-gate before destructive ops
- `command_3-statement-model` — **Front desk** — slash-command wrapper; delegates to 3-statement-model
- `command_competitive-analysis` — **Front desk** — slash-command wrapper; delegates to competitive-analysis
- `command_comps` — **Front desk** — slash-command wrapper; delegates to comps-analysis
- `command_dcf` — **Front desk** — slash-command wrapper; delegates to dcf-model
- `command_debug-model` — **Front desk** — slash-command wrapper; delegates to audit-xls
- `command_lbo` — **Front desk** — slash-command wrapper; delegates to lbo-model
- `command_ppt-template` — **Front desk** — slash-command wrapper; delegates to ppt-template-creator
- `competitive-analysis` — **Drafter** — 2-phase outline-approval gate
- `competitive-analysis_reference_frameworks` — **Foundation** — reference asset
- `competitive-analysis_reference_schemas` — **Foundation** — reference asset
- `comps-analysis` — **Modeler** + **Production** — institutional-grade comps; modeler + slot-filler overlap
- `dcf-model` — **Modeler** — WACC, TV, 75-cell sensitivity table
- `deck-refresh` — **Redliner** — 4-phase approval-gated number swap
- `investment bank-check-deck` — **Diligence** — 4-dimension QC; refuses to edit
- `investment bank-check-deck_reference_ib-terminology` — **Foundation** — reference asset
- `investment bank-check-deck_reference_report-format.` — **Foundation** — reference asset
- `lbo-model` — **Modeler** — 5-section gated workflow; 4-color font convention
- `ppt-template-creator` — **Setup** — meta-skill that creates other skills
- `pptx-author` — **Foundation** — headless .pptx emitter
- `skill-creator` — **Setup** — skill-design teaching reference
- `skill-creator_reference_output-patterns` — **Foundation** — reference asset
- `skill-creator_reference_workflows` — **Foundation** — reference asset
- `xlsx-author` — **Foundation** — headless .xlsx emitter


**fund admin** (6 skills)

- `accrual-schedule` — **Drafter** + **Modeler** — per-accrual derivation + JE; refuse-to-post
- `break-trace` — **Reconciler** — root-cause subtype; single-break trace
- `gl-recon` — **Reconciler** — match-and-classify subtype
- `nav-tieout` — **Reconciler** — independent-recompute subtype
- `roll-forward` — **Reconciler** — foot-check subtype
- `variance-commentary` — **Drafter** + **Triager** — flux commentary with threshold filter


**investment banking** (20 skills)

- `buyer-list` — **Diligence** + **Triager** — parallel scan + tiering
- `command_buyer-list` — **Front desk** — slash-command wrapper; delegates to buyer-list
- `command_confidential infor memorandum` — **Front desk** — slash-command wrapper; delegates to cim builder
- `command_deal-tracker` — **Front desk** — slash-command wrapper; delegates to deal-tracker
- `command_merger-model` — **Front desk** — slash-command wrapper; delegates to merger-model
- `command_one-pager` — **Front desk** — slash-command wrapper; delegates to strip-profile
- `command_process-letter` — **Front desk** — slash-command wrapper; delegates to process-letter
- `command_teaser` — **Front desk** — slash-command wrapper; delegates to teaser
- `confidential info memorandum builder` — **Drafter** + **Production** — CIM builder; narrative + structured artifact
- `datapack-builder` — **Production** — 8-tab xlsx with strict format laws
- `deal-tracker` — **Ledger** — multi-deal pipeline; weekly review mode
- `merger-model` — **Modeler** — accretion/dilution; sources=uses; pro forma EPS bridge
- `pitch-deck` — **Production** — template-first deck with 4 reference sub-skills
- `pitch-deck_reference_calculation-standards` — **Foundation** — reference asset
- `pitch-deck_reference_formatting-standards` — **Foundation** — reference asset
- `pitch-deck_reference_slide-templates` — **Foundation** — reference asset
- `pitch-deck_reference_xml-reference` — **Foundation** — reference asset
- `process-letter` — **Drafter** — IOI / final bid / management meeting invite
- `strip-profile` — **Production** — 1-4 slide quadrant profile; stop-and-confirm per slide
- `teaser` — **Drafter** — anonymous 1-page sell-side teaser; anonymization gate


**operations** (2 skills)

- `kyc-doc-parse` — **Triager** + **Catalog** — parse-and-inventory; structured JSON
- `kyc-rules` — **Rater** — per-rule outcome with citation; routes for human approval


**private equity** (20 skills)

- `ai-readiness` — **Diligence** + **Rater** — cross-portfolio AI scan with go/wait gate
- `command_ai-readiness` — **Front desk** — slash-command wrapper; delegates to ai-readiness
- `command_dd-checklist` — **Front desk** — slash-command wrapper; delegates to dd-checklist
- `command_dd-prep` — **Front desk** — slash-command wrapper; delegates to dd-meeting-prep
- `command_ic-memo` — **Front desk** — slash-command wrapper; delegates to ic-memo
- `command_portfolio` — **Front desk** — slash-command wrapper; delegates to portfolio-monitoring
- `command_returns` — **Front desk** — slash-command wrapper; delegates to returns-analysis
- `command_screen-deal` — **Front desk** — slash-command wrapper; delegates to deal-screening
- `command_source` — **Front desk** — slash-command wrapper; delegates to deal-sourcing
- `command_unit-economics` — **Front desk** — slash-command wrapper; delegates to unit-economics
- `command_value-creation` — **Front desk** — slash-command wrapper; delegates to value-creation-plan
- `dd-checklist` — **Diligence** + **Ledger** — multi-workstream parallel framework with status states
- `dd-meeting-prep` — **Setup** — Socratic question scaffolding
- `deal-screening` — **Triager** — pass / further diligence / hard pass
- `deal-sourcing` — **Researcher** — external discovery -> CRM lookup -> personalized email
- `ic-memo` — **Drafter** — 9-section investment committee memo
- `portfolio-monitoring` — **Monitor** + **Rater** — variance + covenant compliance
- `returns-analysis` — **Modeler** — IRR/MOIC sensitivities
- `unit-economics` — **Diligence** + **Rater** — ARR cohorts, LTV/CAC, retention with quality score
- `value-creation-plan` — **Drafter** + **Setup** — EBITDA bridge, 100-day plan, KPI dashboard


**wealth management** (12 skills)

- `client-report` — **Drafter** — 9-section quarterly performance report
- `client-review` — **Drafter** — review meeting prep
- `command_client-report` — **Front desk** — slash-command wrapper; delegates to client-report
- `command_client-review` — **Front desk** — slash-command wrapper; delegates to client-review
- `command_financial-plan` — **Front desk** — slash-command wrapper; delegates to financial-plan
- `command_proposal` — **Front desk** — slash-command wrapper; delegates to investment-proposal
- `command_rebalance` — **Front desk** — slash-command wrapper; delegates to portfolio-rebalance
- `command_tax-loss-harvesting` — **Front desk** — slash-command wrapper; delegates to tax-loss-harvesting
- `financial-plan` — **Drafter** + **Modeler** — 15-25 page plan with Monte Carlo retirement, scenarios
- `investment-proposal` — **Drafter** — 6-section prospect pitch deck
- `portfolio-rebalance` — **Modeler** — drift vs IPS, tax-aware trades; recommends, doesn't execute
- `tax-loss-harvesting` — **Modeler** + **Diligence** — TLH candidates with wash-sale-safe replacements


---

## Layer 1 — System View

The library as a system, in four zones.

### Zone 1: The Lifecycle

A deal moves through these eight stages. Each carries a dominant archetype; most carry a second.

| # | Stage | Dominant archetype |
|---|-------|-------------------|
| 01 | Source | Researcher |
| 02 | Screen | Triager |
| 03 | Diligence | Diligence + Rater |
| 04 | Model | Modeler |
| 05 | IC Memo | Drafter |
| 06 | Close | Ledger |
| 07 | Value-Create | Drafter + Setup |
| 08 | Monitor | Monitor + Rater |

Eight stages, one library. The work changes with the stage.

### Zone 2: The Foundation

**59 skills · roughly a quarter of the library · the layer the other archetypes run on**

Foundation skills are the shared library. The user never runs them directly — active skills load them as dependencies, the way an analyst reaches for a reference book on the shelf.

Three flagship Foundation utilities loaded byte-identical across plugins:

| Skill | Multiplier | What it does |
|-------|-----------|--------------|
| `audit-xls` | × 6+ agents | spreadsheet audit · reports first, fixes on request |
| `xlsx-author` | × 8+ agents | writes the Excel outputs in headless sessions |
| `pptx-author` | × 3+ agents | writes the PowerPoint outputs in headless sessions |

The remaining ~50+ Foundation files are reference-flavored: formula libraries, formatting standards, vendor entity dictionaries, query plan templates, accounting policy snippets, rules grids, scoring rubrics, deck templates.

### Zone 3: The Practice Domains

The library is organized around 3 top-level domains with 20+ sub-groupings.

| Domain | Count | Sub-groups |
|--------|-------|-----------|
| **Agent plugins** | 70 skills · 10 agents | earnings reviewer, gl reconciler, kyc screener, market researcher, meeting prep agent, model builder, month end closer, pitch agent, statement auditor, valuation reviewer |
| **Partner built** | 25 skills | LSEG (16) · S&P Global (9) |
| **Vertical plugins** | 118 skills · 7 verticals | equity research, financial analysis, fund admin, investment banking, operations, private equity, wealth management |

### Zone 4: The 17 Active Archetypes

Family-grouped vocabulary:

- **Navigators:** Front desk · Orchestrator · Gatekeeper · Vendor pipeline · Setup
- **Gatherers:** Researcher · Monitor · Ledger · Catalog
- **Analyzers:** Modeler / Calculator · Reconciler · Rater / Scorer · Triager · Diligence
- **Makers:** Drafter · Redliner · Production

---

## Layer 2 — The Skeleton

Ten architectural moves a skill.md can make, in conceptual order. Not every skill makes all ten — some embed refusals inside workflow gates, some skip the matter-context check entirely. The vocabulary stays consistent across the ones that do.

| # | Move | What it carries |
|---|------|-----------------|
| 1 | **Header** | name · description · argument-hint · user-invocable — how the skill is identified and invoked. Some skills also declare a `tools` list to grant MCP access. |
| 2 | **Purpose** | What the skill does, and why, in one paragraph. |
| 3 | **Matter context** | The data trust boundary — trusted firm sources (rules grids, NAV packs, accounting policy) vs. untrusted external docs (subledger extracts, vendor statements, transcripts). |
| 4 | **Hard gates** | Conditions the skill checks before doing anything. If a precondition isn't met, the skill stops before any further work. |
| 5 | **Load playbook** | Read the firm's CLAUDE.md, rules grid, or accounting policy for team-specific positions. The skill compares; the firm's playbook decides. |
| 6 | **Workflow** | Numbered steps. The body. Four to six typical; each with branching or output rules. Finance workflows often gate at every step. |
| 7 | **Output formats** | Memo · slack · tracked-change docx · xlsx workbook · pptx deck · structured JSON disposition. Varies by role and by destination. |
| 8 | **Consequential-action gate** | Pause before high-stakes commits — posting (journal entries), sending (client deliverables), executing (trades), publishing (research), approving (KYC), distributing (Limited Partner reports). Requires explicit yes to proceed. |
| 9 | **What it does NOT do** | The refusals. Some skills name explicit boundaries as a labeled section. Many skills embed the same boundaries within workflow gates or critical-warning blocks instead. The placement varies; the refusal is still part of the architecture. |
| 10 | **Closing action** | Next-steps decision tree. The skill ends by handing the user choices. |

---

## Layer 3 — The Seventeen Archetypes

The active library, grouped into four families of work.

### Family 1 — Navigators · Traffic & Boundaries

Manage the flow of intent — route requests, gate access, coordinate over time, scaffold the user's own work. They never do the substantive analysis or production themselves.

#### 01 · Front desk

*A thin slash-command at the front of the room. `argument-hint` frontmatter; takes the order, runs small adapter logic, hands the job to a named domain skill.*

- **Examples:** command_dcf · command_ic-memo · command_screen-deal · /analyze-bond-basis · /analyze-fx-carry
- **Count:** ~46 · 22% of the library
- **Signature row:** Header (with argument-hint) lights up; everything else stays thin

#### 02 · Orchestrator

*The general contractor. Breaks a complex task down and manages multiple sub-skills.*

- **Examples:** earnings-reviewer · gl-reconciler · kyc-screener · pitch-agent · model-builder · statement-auditor · valuation-reviewer · tear-sheet
- **Count:** ~10 · one per agent plugin
- **Often pairs with:** Foundation · Production (tear-sheet)
- **Signature rows:** Workflow + Output formats

#### 03 · Gatekeeper

*Checks the room and explicitly refuses to let the AI proceed if prerequisites are missing.*

- **Examples:** initiating-coverage (5-task sequence with one gate per task)
- **Count:** ~1 primary · appears as secondary in larger workflows
- **Often pairs with:** Diligence · Monitor
- **Signature row:** Hard gates (gatekeeping check)

#### 04 · Vendor pipeline

*A vendor-integration shape. "Let the tools compute, you interpret" — the discipline rule.*

- **Examples:** bond-futures-basis · bond-relative-value · equity-research · fixed-income-portfolio · fx-carry-trade · macro-rates-monitor · option-vol-analysis · swap-curve-strategy
- **Count:** 8 · all LSEG
- **Often pairs with:** Drafter · Diligence
- **Signature rows:** Workflow (vendor MCP tool chain) + Output formats (mandated tables + directional call)

#### 05 · Setup

*Guides work; refuses to substitute. Hard "no rewriting" rule. Includes meta-creators that set up by withholding the artifact.*

- **Examples:** skill-creator · ppt-template-creator · dd-meeting-prep · value-creation-plan
- **Count:** ~3-4
- **Often pairs with:** Drafter (value-creation-plan)
- **Signature rows:** Hard gates + Workflow (guide the user) + What it does NOT do (rewrite)

---

### Family 2 — Gatherers · Ingestion & Memory

Bring the outside world into the agent's context — external data, ongoing state, reference inventory. The eyes and ears.

#### 06 · Researcher

*Pulls external data and synthesizes it, with strict attribution. Primary value is the gathering. Date-freshness gates ("training data is outdated, search for latest") are a Researcher signature.*

- **Examples:** sector-overview · deal-sourcing · funding-digest · spglobal earnings-preview-beta
- **Count:** ~7
- **Often pairs with:** Drafter · Vendor pipeline · Diligence
- **Signature rows:** Hard gates (date freshness) + Workflow (gather + synthesize) + Output formats

#### 07 · Monitor

*The persistent observer. Watches state for drift, sweeps for upcoming events. Event-driven.*

- **Examples:** portfolio-monitoring · catalyst-calendar · morning-note
- **Count:** ~3
- **Often pairs with:** Triager · Rater · Diligence
- **Signature rows:** Hard gates + Workflow (sweep)

#### 08 · Ledger

*Maintains the structured state of a project across its lifecycle — init / update / review / close modes.*

- **Examples:** deal-tracker · thesis-tracker · initiating-coverage (workflow tracking)
- **Count:** ~3
- **Often pairs with:** Triager · Monitor · Diligence
- **Signature rows:** Workflow (multi-mode) + Closing action (route to next mode)

#### 09 · Catalog

*Tabular output with every cell cited. Refuses paraphrase. Structured extraction from packets.*

- **Examples:** kyc-doc-parse · datapack-builder · entity dictionaries
- **Count:** ~3
- **Often pairs with:** Foundation · Production · Pedagogical reference
- **Signature row:** Output formats (citation per cell)

---

### Family 3 — Analyzers · Logic & Rules

Apply strict computational or evaluative frameworks. They calculate and recommend; they defer the final decision to a human.

#### 10 · Modeler / Calculator

*Quantitative computation against templates with formulas-over-hardcodes discipline, section-by-section verification gates, integrity checks (BS balance, cash tie-out, sensitivity discipline). The artifact is a live workbook.*

- **Examples:** 3-statement-model · dcf-model · lbo-model · merger-model · returns-analysis · accrual-schedule · portfolio-rebalance · tax-loss-harvesting
- **Count:** ~14
- **Often pairs with:** Production · Foundation · Drafter (financial-plan)
- **Signature rows:** Hard gates (template required) + Workflow (verification per section) + Output formats (live workbook)

#### 11 · Reconciler

*Matches two-sided data and diagnoses breaks, but refuses to post the final adjustment.*

- **Examples:** gl-recon (match-and-classify) · break-trace (root-cause) · nav-tieout (independent-recompute) · roll-forward (foot-check)
- **Count:** ~8 · 4 observed subtypes
- **Often pairs with:** Foundation · Production (tear-sheet)
- **Signature rows:** Hard gates (tolerance) + Workflow (compare + classify) + What it does NOT do (post adjustments)

#### 12 · Rater / Scorer

*Applies a written rubric (like KYC rules) to score an input, forcing a human to sign off on the disposition. Each outcome cites a rule.*

- **Examples:** kyc-rules · unit-economics scoring · ai-readiness · portfolio-monitoring quality score
- **Count:** ~2 primary · pairs frequently as secondary
- **Often pairs with:** Triager · Monitor · Diligence
- **Signature rows:** Load playbook (rubric) + Workflow (per-rule outcome) + Output (citation per outcome)

#### 13 · Triager

*Sorts complex inputs into clear buckets (e.g., pass/fail/diligence) based on a playbook.*

- **Examples:** deal-screening (pass / further diligence / hard pass) · idea-generation (5 named screens with ranked shortlist)
- **Count:** ~5
- **Often pairs with:** Production · Monitor
- **Signature rows:** Load playbook (criteria) + Workflow (bucket decision) + Closing action

#### 14 · Diligence

*Runs multiple frameworks in parallel; flags gaps with source attribution. The heaviest analyzer.*

- **Examples:** dd-checklist · ib-check-deck · unit-economics · ai-readiness · audit-xls (variant) · buyer-list
- **Count:** ~13
- **Often pairs with:** Foundation · Ledger · Gatekeeper · Monitor
- **Signature rows:** Workflow (parallel frameworks) + Output formats (per-workstream findings + gaps)

---

### Family 4 — Makers · Artifact Generation

Produce the actual deliverables — memos, models, edits, decks. Bound by the law of layout; they do the typing, the human does the sending.

#### 15 · Drafter

*Produces a complete narrative, but carries a hard refusal to press "send".*

- **Examples:** ic-memo · earnings-analysis · earnings-preview · morning-note · sector-overview · competitive-analysis · investment-proposal · client-report · client-review · financial-plan · cim-builder · process-letter · teaser · accrual-schedule · variance-commentary
- **Count:** ~25 · largest Maker archetype
- **Often pairs with:** Production · Researcher · Modeler · Setup
- **Signature rows:** Workflow (narrative build) + Output formats (memo / docx) + What it does NOT do (transmit)

#### 16 · Redliner

*Edits an existing artifact but is strictly barred from deleting sections or weakening guardrails. A revision adds a new row; prior rows stay.*

- **Examples:** model-update · clean-data-xls · deck-refresh · customize
- **Count:** ~5
- **Often pairs with:** Modeler · Production
- **Signature rows:** Hard gates (proposal before destructive ops) + Workflow (single-section edit) + What it does NOT do (delete sections)

#### 17 · Production

*Slot-filler for structured artifacts (xlsx · pptx · docx). Layout law dominates; the artifact IS the deliverable.*

- **Examples:** pitch-deck · comps-analysis · lbo-model · datapack-builder · strip-profile · client-report · xlsx-author · pptx-author
- **Count:** ~9 primary · ~18 with secondary
- **Often pairs with:** Drafter · Modeler · Foundation
- **Signature rows:** Hard gates (template required) + Load playbook (format laws) + Workflow (slot-fill + verify) + Output formats (final file) + What it does NOT do (rebuild layout / invent slots)

---

## Closing — Three Questions Every Reliable Skill Answers

Read the catalog long enough and three questions surface. They sit underneath every skill.md in the library — and underneath every reliable AI workflow you'll meet. Read with them; write with them.

### 1. What does the skill refuse to do?

Every skill.md carries a *What this does NOT do* block, explicitly and sometimes implicitly. The refusals name where the skill's authority ends.

### 2. What conditions are checked before any work begins?

Every reliable skill tests its preconditions first. If the conditions fail, the skill halts — a Reconciler that exits when totals don't tie, a Modeler that exits without a template, an Orchestrator that exits when sources can't be trusted. The gate names the universe this skill lives in, and the universe where it's the wrong tool.

### 3. Where does the skill hand the call back to a human?

Every reliable skill ends with a deference move. The verb varies — *post* a journal entry, *send* a client deliverable, *execute* a trade, *publish* a research note, *approve* a KYC packet, *distribute* a Limited Partner report. Spot the verb and you spot the trust boundary. The skill's authority ends there; yours begins.

---

## Part of Cartu Collection

*Cartu* — from the Latin *charta*, a chart. The act of translating dense text into a drawn shape.

*Claude Finance* is a *Cartu* reading. It takes public markdown files and draws the shape of each skill — to increase comprehension. Wander a bigger collection at [cartu.app](https://cartu.app).

## Open Access

*Cartu / Claude Finance* is part of the Cartu collection. The library is a collection of architectural shape annotations drawn from Anthropic's `financial-services` skill.md library — 213 skills across 3 practice domains, released May 2026. A gift built to pay forward.

## Sources & Attribution

The public skill.md repository library is drawn from [`anthropics/financial-services`](https://github.com/anthropics/financial-services) under Apache 2.0, released May 2026. Every skill.md carries a link back to its source file. Original analysis (the four families, the seventeen active archetypes, the Foundation layer, the lifecycle zone, the diagrams) by T Ngo / HelloTNgo.

*The Shape of a Skill* is built by T Ngo. Thought partner with Claude Code Opus 4.7. Shipped with Lovable.

---

*Live at [claudefinance.cartu.app](https://claudefinance.cartu.app) · Source: [github.com/thienthy8/cartu-claudefinance.app](https://github.com/thienthy8/cartu-claudefinance.app)*