Case Study Orchestrator

Case Study Orchestrator

A Claude Code plugin that solves financial-services case studies end-to-end. Hand it a folder of case materials (a prompt PDF, source documents, model templates), invoke /solve-case-study, and it produces the deliverables — typically a filled .xlsx, a .pptx, or both — to <case-folder>/_output/.

What's new in v0.2.0

Based on a real-world Summit LBO case run (May 2026):

• Native Excel two-input data tables for sensitivity grids. The modeler now writes DataTableFormula (via the bundled scripts/sensitivity_datatable.py) instead of hardcoded numeric IRR grids — closes the single material gap between the plugin's first deliverable and what shipped to the hiring manager.
• LBO templating-convention audit (scripts/audit_lbo_template.py). Extends financial-analysis:audit-xls with checks the upstream skill does not perform: hold-flag gating per operating row, one-time S&U row spreading with acq/exit-flag gating, debt amort MIN(amort, opening_balance) cap, terminal closing-debt = 0, sensitivity-table format, in-workbook assumption labels.
• Protected-sheet preservation (scripts/preserve_protected_sheets.py). Copies reference sheets (Answer Key, Variance) byte-for-byte instead of letting openpyxl re-encode and drift float precision.
• Label-vs-spec conflict surface. When a row label conflicts with prompt text (e.g., label says "Quarterly Rate" but spec says "Annual"), the modeler surfaces the conflict to the orchestrator rather than picking silently.
• No autonomous v2 rewrites. The modeler's first deliverable is the canonical answer; v2 happens only on explicit user direction. Working scripts go to <case-folder>/_output/_artifacts/, not alongside the deliverable.
• Tightened dispatch-prompt discipline. The orchestrator hands workers paths + short skill hints, not 150-line formula prescriptions.
• Versioning convention clarified. Subsequent revisions: .FILLED.v2.xlsx (dot-separated before .xlsx, not underscore-embedded).

Architecture

One orchestrator skill (solve-case-study) plus three worker subagents (case-digester, case-modeler, case-deck-builder). The orchestrator inventories the case folder, classifies it on three flags (has_narrative_source_docs, has_model_template, requires_deck), and dispatches the smallest set of workers required.

Case shape	Workers dispatched	Example
Model-only	case-modeler	LBO build from a trial balance
Deck-only with narrative sources	case-digester → case-deck-builder	2-slide transaction summary from source PDFs
Model + deck (deck depends on model)	case-modeler → review gate → case-deck-builder	LBO + 2-page opportunity overview
Parallel sub-exercises	case-digester ∥ case-modeler → case-deck-builder	Pack with both a narrative-deck task and a standalone modeling task

Workers run in parallel when their inputs are independent. A single review gate fires before the deck-builder only when the deck depends on prior worker output.

Routing through Anthropic's plugins

All methodology comes from skills preloaded into the workers via the skills: frontmatter field — full skill content is injected at subagent startup, not just the description.

• case-modeler preloads: financial-analysis:3-statement-model, lbo-model, dcf-model, comps-analysis, clean-data-xls, audit-xls, xlsx-author, private-equity:returns-analysis.
• case-deck-builder preloads: investment-banking:pitch-deck, strip-profile, datapack-builder, financial-analysis:ib-check-deck, pptx-author, private-equity:deal-screening, dd-checklist, ic-memo.
• case-digester preloads nothing — it is a focused read-only synthesizer.

The plugin adds only what Anthropic's pre-built model-builder and pitch-agent subagents don't ship: Bash access (so the bundled python scripts in xlsx-author / pptx-author can run) and case-study-specific I/O (reading prompts from PDFs or in-Excel instructions, handling yellow/blue template conventions, etc.). Anthropic's *-agent plugins are wired for live-Excel mode via mcp__office__excel_* and don't include Bash; they don't fit headless case-study workflows out of the box.

Required plugins

These must be enabled for the worker preloads to resolve:

• financial-analysis@claude-for-financial-services (v0.1.1+)
• investment-banking@claude-for-financial-services (v0.2.1+)
• private-equity@claude-for-financial-services (v0.1.2+)

If a preloaded skill is missing, Claude Code skips it silently and writes a warning to the debug log. The plugin still functions in degraded form but loses the corresponding methodology.

Install

/plugin marketplace add ZaBrisket/case-study-orchestrator
/plugin install case-study-orchestrator@case-study-marketplace

Usage

cd path/to/case-folder
/solve-case-study

Or with an explicit folder:

/solve-case-study /absolute/path/to/case-folder

The orchestrator inventories the folder, classifies the case, and shows you a one-line dispatch plan before any subagent fires. A review gate fires before the deck-builder only when the deck depends on prior worker output.

Output

Deliverables write to <case-folder>/_output/. Original input files are never modified. Each run suffixes .v2, .v3 etc. rather than overwriting existing deliverables.

What's deliberately not included

• No bundled MCP, no userConfig, no hooks. Self-contained and headless.
• No external research in the default flow. Case studies typically forbid it. Invoke firecrawl skills separately between waves if a case explicitly allows external sources.
• No cross-case memory. Each dispatch is a fresh classification.
• No dispatch to Anthropic's model-builder or pitch-agent subagents. Verified to be wired for live-Excel mode with no Bash access; they would fail to run the bundled python in headless case-study runs.

Versioning

plugin.json pins version: "0.1.0". Per Claude Code's plugin-caching rules, the cache key is derived from this field. Bump the version on every release; otherwise /plugin update reports "already at latest" even after new commits are pushed.

Files

case-study-orchestrator/
├── .claude-plugin/
│   └── plugin.json
├── README.md
├── skills/
│   └── solve-case-study/
│       └── SKILL.md                     ← orchestrator (user-invocable, /solve-case-study)
├── agents/
│   ├── case-digester.md                 ← read-only worker
│   ├── case-modeler.md                  ← writes .xlsx; preloads modeling skills
│   └── case-deck-builder.md             ← writes .pptx; preloads deck + screening skills
└── scripts/                             ← bundled helpers (v0.2.0+)
    ├── audit_lbo_template.py            ← LBO templating-convention audit
    ├── sensitivity_datatable.py         ← native two-input data tables
    └── preserve_protected_sheets.py     ← byte-for-byte sheet preservation