Portable Agent Skills library that helps an investor study a listed company as an owner and operator. It uses role-based, Socratic boardroom conversations to connect operational choices to future free cash flow (FCF) — instead of producing generic investment reports or trading recommendations.
- Understand how a company creates, consumes, and allocates cash.
- Reason as a board director or controlling owner, not as a short-term trader.
- Separate verified facts from management claims, inferences, modelling assumptions, and unknowns.
- Convert departmental insights into a small, testable set of FCF drivers, scenarios, and counterevidence.
There are two ways to install, two philosophies:
- Skills CLI copies the skills into your project so you can edit them and make them your own.
- Claude Code plugin keeps them as a managed, read-only, always-current bundle you subscribe to rather than fork.
npx skills@latest add white-source/value-grill-skillWorks with Claude Code, Codex, Cursor, Copilot, and other Agent-Skills-compatible harnesses. Files are copied into your repo and are yours to modify.
Prefer a plug-and-play install you don't maintain by hand? This skill set also ships as a native Claude Code plugin. Instead of copying editable files into your repo, the plugin installs the whole set as a managed bundle that updates when a new version ships.
Inside Claude Code:
/plugin marketplace add white-source/value-grill-skill
/plugin install value-grill-skill@white-source
Or from your shell:
claude plugin marketplace add white-source/value-grill-skill
claude plugin install value-grill-skill@white-sourceNo separate setup step is needed — start a session with /value-boardroom [company] right away.
The fourteen single-role shortcuts (/value-role-secretary, /value-role-ceo, … /value-role-ir-pr) are plain slash-command files in ./commands/. Each harness discovers slash commands from a different place, so availability depends on how you installed:
| Harness | How /value-role-* appears |
|---|---|
| Claude Code | Automatic if installed as a plugin (Option 2) — the plugin's root commands/ is read on load. Or run scripts/install-commands.sh (Option 1). |
| ZCode | Same as Claude Code (plugin root commands/), since ZCode is Claude-Code compatible. |
| Kilo | Run scripts/install-commands.sh once to link the commands into ~/.config/kilo/command/. |
| Codex | Each role ships as its own SKILL.md (in skills/value-role-*/), so $value-role-ceo, $value-role-cfo, … work natively after npx skills add. |
If you installed via the Skills CLI (Option 1) rather than the plugin, the role shortcuts work everywhere automatically: in Codex they appear as $value-role-* skills, and in Claude Code / ZCode / Kilo you can also wire them as /value-role-* slash commands with the bundled helper:
scripts/install-commands.sh # symlink into Claude Code, ZCode, Kilo
scripts/install-commands.sh --copy # copy instead of symlink (e.g. for Windows)
scripts/install-commands.sh --uninstall # remove them again
scripts/install-commands.sh --target kilo --target claude # restrict to given harnessesReload / restart your harness after running it.
/value-boardroom [company name or ticker]
On start the session confirms the company (and identifier/market if known), the as-of date, the research aim, and the evidence mode. The role mode is just one quick question (default briefing) — no flags to type.
This library's signature feature is role reversal: you can study a company from either side of the boardroom table, and the same fourteen operating roles work in both directions.
| Mode | Model plays | You play | Why use it |
|---|---|---|---|
briefing (default) |
The selected executive — with their incentives, budgets, and conflicts | A board director / partner asking questions | Build understanding by cross-examining real operating detail. The executive must justify every number all the way to cash flow. |
examination |
The board | The CEO, CFO, or selected executive — answering | Pressure-test whether you actually understand the business. The board catches the gaps and hand-waving in your own reasoning. |
briefing = learn by interrogating. examination = learn by being interrogated. You pick the mode with one quick question at startup (default briefing), and switch any time with /value-mode — or just say it ("reverse roles", "考考我"). No restart needed. Either way, the turn contract is identical: state a fact or labelled assumption, expose one incentive conflict, trace a clear cash-flow path, and ask exactly one question.
Fourteen roles cover the whole company. Enter one with the auto-completing /value-role-<role> shortcut (e.g. /value-role-cfo), chain several with /value-role ceo cfo ..., or step through with /value-next. Each speaks only within its remit, surfaces a budget or strategic conflict, and traces its point to revenue, cost, working capital, CapEx, or capital allocation.
| Role | What they own | What to interrogate |
|---|---|---|
secretary |
Board secretary / coach — agenda, question tree, evidence, assumptions | Whether every conclusion is sourced and traceable |
ceo |
CEO — business model, growth sources, strategy, culture, capital allocation | Business model defensibility (lock-in, switching cost, network effects), organic vs acquired growth, reinvestment rate, culture-talent alignment, returns on capital |
cfo |
CFO — profit quality, leverage, working capital, allocation | Cash conversion cycle, accrual vs cash profit, maintenance vs growth CapEx |
product |
Product / R&D — customer problem, roadmap, substitution risk | Pricing power, retention/churn, R&D efficiency |
marketing |
Marketing / brand — acquisition, loyalty, ad dependency | CAC, LTV, what declines first if advertising stops |
sales |
Sales — order quality, pricing, concentration | Realized price after discounts, customer concentration, receivables |
channel |
Channel — distributor inventory, merchandising, rebates | Channel inventory days, sell-through, channel margin |
supply |
Supply chain / procurement — input costs, supplier power | COGS supplier concentration, contract terms, lead times |
operations |
Operations / manufacturing — utilization, yield, unit cost | Capacity utilization, fixed-cost absorption, expansion CapEx |
customer-success |
Customer success / after-sales — retention, churn, renewals | Gross/net retention, expansion vs contraction, service cost |
people |
People / org — critical talent, incentives, succession | Attrition in key roles, incentive structure, bottlenecks |
legal-risk |
Legal / compliance / risk — licences, litigation, regulation | Contingent liabilities, regulatory exposure, data governance |
strategy-ma |
Strategy / M&A — new lines, expansion, acquisitions | Post-acquisition returns, goodwill, integration cost |
ir-pr |
IR / PR — commitments, transparency, reputation | Public guidance vs actuals, narrative vs operations |
| Command | What it does |
|---|---|
/value-ask [question or goal] |
Describe what you want; get the right command with an explanation |
/value-boardroom [company] |
Initialize or reset a company boardroom session |
/value-mode [briefing|examination] |
Switch role mode anytime — or just say "reverse roles" |
/value-role-<role> |
Single-role shortcut that auto-completes — type /value-role- and pick (e.g. /value-role-cfo) |
/value-role [role...] |
Chain several executives in sequence, e.g. /value-role ceo cfo supply |
/value-next |
Step to the next role on the suggested tour — no retyping |
/value-research [topic] |
Collect or request verifiable public information and classify it by evidence class |
/value-assumption |
View or revise material modelling assumptions |
/value-evidence |
Show the evidence ledger |
/value-cashflow |
Translate operating results into revenue, margin, working capital, CapEx, FCF, and ROIC |
/value-stress-test [shock] |
Run a named shock or build bear/base/bull cases through the cash-flow chain |
/value-kill [year] |
Pre-mortem: assume cash-flow failure by the chosen year and work backwards |
/value-board-summary |
Exit roleplay and synthesize a traceable board conclusion |
Not investment advice. Every claim carries an evidence class. Full walkthrough (Chinese) →
A condensed run of the main commands on a real listed company, with real sourced financials (Yahoo Finance TTM, Wikipedia). Every number is tagged by evidence class.
> /value-boardroom Pop Mart --mode briefing
Company confirmed. Pop Mart International (HKEX: 9992), as-of July 2026. Briefing mode.
> /value-role-cfo
| Metric | TTM (HKD) | Class |
|---|---|---|
| Revenue | 37.1B | Fact |
| Net income | 12.8B (margin 34%) | Fact |
| Levered FCF | 9.1B | Fact |
| Cash | 17.2B (D/E 12.6%) | Fact |
| ROE | 77.6% | Fact |
FCF/net ≈ 71% → ~29% of profit absorbed by working capital + CapEx (Inference). Light asset, high margin, low debt.
> /value-cashflow
Avg price ¥59–69/box (Fact, Wikipedia) × units × IP mix → revenue ¥34B (Fact, Yahoo Finance TTM) → high margin → light WC (cash at sale) → CapEx (stores/roboshops/overseas) → FCF HKD 9.1B TTM. Three drivers: same-store traffic, overseas ROIC, premium-category mix (Inference).
> /value-stress-test price -10%
Chain: price ↓ → volume? (Unknown) → fixed cost unchanged → operating leverage worsens → FCF pressured. Bear/Base/Bull with falsification criteria, not a single forecast.
> /value-kill 2030
Pre-mortem: Labubu fatigue → comps weaken → overseas payback stretches >36 months → growth CapEx becomes a cash drain → 2030 crunch. Early indicator: overseas comps 3-quarter decline + payback >3yr.
> /value-board-summary
Six-section conclusion: FCF mechanism tracked to sourced data, 5/10/20yr stability assessment, conditional scenarios, ≤3 drivers, counterevidence ranked by materiality, next agenda with roles + evidence + metrics to observe.
Not investment advice. No target price, no buy/sell/hold.
Every claim is labelled so conclusions stay traceable:
| Class | Meaning |
|---|---|
Fact |
A source-backed claim, including publication date and source |
Management statement |
A company claim not independently verified |
Inference |
An operating conclusion derived from cited facts |
Assumption |
A model input that could be falsified |
Unknown |
Material information without reliable support |
Public-web research is used when available and source-dated. When research tools are unavailable, the skills degrade honestly: they state the missing evidence, name the primary sources to consult, suggest searches, and pose the next board question. They never invent data, sources, access, or verification.
This is a research and learning framework. It does not provide investment recommendations, target prices, buy/sell/hold calls, or trade instructions.
npm testRuns the Node built-in test runner (node --test tests/skills.test.mjs), which validates the skill layout, frontmatter, shared references, and safety contracts. Requires Node 20+.
MIT