Equity Initiation Report

Initiating Coverage

A first-time coverage report: the argument for or against owning a company, built from original research, a full model and a defended valuation. Five tasks, one per request by default, each verifying its inputs before it starts. The deliverables are a 30 to 50 page DOCX and an XLSX model, with typography set by .agents/skills/docx/SKILL.md.

Evidence labels, source tiers, staleness, the readiness posture and the intake limits: .agents/skills/research-conventions/SKILL.md, read before the first deliverable.

One task per request

This skill runs one task per request by default: execute the task, deliver its file, and stop, so the user can review before the next.

That stop is an intake exception and yields under .agents/skills/research-conventions/references/intake.md. When the user asked for the whole pipeline in one request ("do all five tasks", "run the whole initiation") or this skill is running as an input to another workflow, run the tasks in order without stopping between them, naming at each handoff the task just finished.

When the request names no task and does not ask for the whole run ("write an initiation report for X"), reply with the task list and ask which one to begin, in roughly this shape:

An initiation for [Company] runs as five tasks, one per request:

1. Company research      4. Chart generation
2. Financial modelling   5. Report assembly
3. Valuation analysis

Each one delivers its file and stops, so you can review before the next.
Which would you like to start with?

Either way, never assume a task the user did not ask for, and never start a task whose prerequisites are unverified.

Task map

Tasks 1 and 2 are independent of each other and can run in either order. Tasks 3, 4 and 5 have hard dependencies: verify the named inputs exist and can be opened, and when one is missing, stop and say which task supplies it rather than building a placeholder.

Workbook manifest

Tasks 2 and 3 build one file, [Company]_Financial_Model_[Date].xlsx. Its sheets, in this order, are enumerated here and pointed at from everywhere else:

Cover, Revenue Model, Income Statement, Cash Flow Statement, Balance Sheet, Scenarios, DCF Inputs, DCF, Sensitivity Analysis, Comparable Companies, Precedent Transactions (optional), Valuation Summary, Checks.

Task 2 builds Cover, the six core operating tabs Revenue Model through DCF Inputs, and Checks. Task 3 adds the four valuation tabs DCF, Sensitivity Analysis, Comparable Companies and Valuation Summary, adds Precedent Transactions where the deal set supports one, and extends Cover and Checks to cover them. Twelve sheets, thirteen with Precedent Transactions. What Cover and Checks each hold, and why the file opens on one and closes on the other, are in .agents/skills/xlsx/SKILL.md.

Standing rules

These bind every task.

Deliver one file per task, plus the delivery message Task 5 specifies, and no additional files. No standalone executive summary, no quick-reference guide, no next-steps document. Extras cost context and are not part of the workflow.

Never overwrite the user's own file. When the user supplied a model or a draft, produce a versioned copy or an additive companion, and mark what changed.

The minimums are floors, not targets. Word counts, chart counts and table counts below are the point at which a section stops being a stub. They are met by analysis, never by padding, and assets/quality-checklist.md says where the words go and what to cut.

Every material claim carries its evidence label and its source. Labels, source tiers, the conflict protocol and the staleness thresholds live in .agents/skills/research-conventions/references/evidence.md. Keep a source register as you work, one row per source: id, name, type, publication date, access date, period covered, location inside the document, tier, and whether it is stale. The register becomes the report's data-sources page, so building it during research costs nothing and rebuilding it at the end costs a day.

Evidence confidence and underwriting status are two different fields. High confidence in the numbers does not mean the position can be underwritten: the numbers can be excellent and the capital structure, the valuation or the disclosure still leave ownership unsupportable. Both appear on the report cover, and neither substitutes for the other.

Step 0: Classify the initiation

Before Task 1, name the mode. It decides which content is required, how long each section runs and what tone the argument takes: a buy-side deep dive is short on boilerplate and long on falsifiability, a long-only initiation centres durability and capital allocation, a sector launch maps the battlefield before naming winners.

The eight modes, what each one requires and what each one cuts: references/report-modes.md. Read it once at Step 0 and carry the mode through every later task.

Complete when the mode is written down in one line with the reader it serves, and every later task's content list has been checked against it.

Task 1: Company research

Business, management, competitive position, industry, TAM and risks. Detailed workflow in references/task1-company-research.md; the shape a thesis pillar, a debate and a risk row must take is in references/argument-standards.md.

Thin context. When the company is too thinly covered to research properly (a recent listing, a foreign private issuer with sparse English disclosure, a carve-out with no standalone history), do not fill the gap with plausible detail. Deliver a skeleton instead: the research agenda, the model architecture the company would need, the specific source requests that would unblock it, and the thesis hypotheses to be tested. Say plainly that it is a skeleton and what would upgrade it.

Complete when the document runs 6,000 to 8,000 words and carries: the company overview and history; 300 to 400 word bios for each of 3 to 4 executives; products and services; the industry; 5 to 10 competitors analysed rather than listed; TAM with its build; 8 to 12 risks across the four categories; and three to five thesis pillars each with the four parts required by references/argument-standards.md.

A skeleton completes instead when it carries all four items named in the thin-context paragraph above, says on its first page that it is a skeleton and what would upgrade it, and contains no invented detail. The word count, the bios and the competitor and risk counts do not apply to it.

Task 2: Financial modelling

Extract the historicals, then build the projection model. Detailed workflow in references/task2-financial-modeling.md. Workbook construction rules, live formulas and the Checks sheet belong to .agents/skills/xlsx/SKILL.md.

Before starting, confirm one of: access to the 10-K or the statements; or historical financials the user supplied covering income statement, cash flow and balance sheet for 3 to 5 years.

Complete when the workbook opens and holds every Task 2 sheet of the workbook manifest, with 3 to 5 years of history and 5 years of projections; the revenue model carries 20 to 30 product rows and 15 to 20 geography rows; the income statement carries 40 to 50 line items; and all three scenarios differ in their parameters rather than in their labels.