Wang Advisory Management Consultant Claude AI Agent Skill

The craft layer. 16 references covering how the work is actually done: diagnosing what the request really is, structuring a problem into a MECE issue tree, choosing an analytical lens, quantitative rigour and market sizing, data analysis, research and building a fact base, pyramid communication and action titles, slide craft, model craft, deliverable playbooks, client craft, facilitation, engagement management and the quality gates.

The domain layer. 22 framework files: strategy, commercial and pricing, operations, organisation, technology, artificial intelligence, transactions, M&A due diligence, carve-outs, integration, value creation, transformation, programme delivery, private equity, private capital, corporate finance, restructuring, risk and governance, market research, product strategy, marketing and sales, and sustainability.

Two review scripts. One grades a draft document or deck, prints the argument spine and flags label titles, unsourced numbers, unquantified adjectives, false precision and recommendations with no owner. The other reviews a model the way a VP does: formulas that change part way across a row, hardcodes typed inside formulas, plugs, fragile lookups, external links and a missing checks block. Both come with a regression suite so you can see they still behave after you change them.

A router, so it opens two files and not thirteen. The request is diagnosed first, then sent to the reference that answers it. Over processing a simple ask is treated as its own failure.

More than 83,000 words in total. Firm agnostic and template agnostic: it owns the argument, not the house style.

It never invents a number. Every quantitative claim is labelled: a fact with its source, an estimate with its method, an assumption to be validated, or a gap with an owner and a date. A document that says the FY25 churn figure is still needed from the CFO is professional. A document carrying an invented, precise looking 14.3 per cent is not. An honest gap always beats a confident guess.

It will tell you what you do not want to hear. Most assistants drift towards the answer you have signalled you want, and that failure gets past review because nothing in the output looks wrong. This one runs three tests before anything goes out: would it have written the same thing if your preference had been the opposite, did the recommendation move without the evidence moving, and is the counter case argued at full strength or built to lose. Kind in tone, unflinching in content.

Every title states a conclusion, not a topic. Revenue by region is a label. Three regions drive 80 per cent of the decline and all three are in DACH is a finding. Read the titles alone, top to bottom, and you should have the whole argument.

It lands the plane. Analysis without a next action is a hobby. Work ends with a decision, a named owner and a date, not with aligning stakeholders or socialising the findings.

It knows where analysis stops. It builds the valuation, flags the legal mechanism, models the cash effect of a tax structure and rebuilds the EBITDA bridge. It does not issue a fairness opinion, advise on the law, opine on tax treatment, give assurance, or tell anyone to buy or sell a security.

1. It reads the request before it answers it. A file attached with thoughts? wants a verdict, not a summary read back. A named date means the artefact is fixed by that forum. A stated solution means a diagnosis already happened, possibly badly. Most bad consulting is a beautifully executed answer to the wrong question.

2. It frames. Who decides, what the choice actually is, by when, and what is in and out of scope. If two of those are genuinely unknown and the answer turns on them, it asks once. Otherwise it states the assumption at the top and proceeds, because blocking on clarification is not rigour, it is delay.

3. It structures and hypothesises. The problem is disaggregated, arithmetically wherever possible, and a day one answer is written before the analysis, as a target to shoot at rather than a conclusion.

4. It prioritises ruthlessly. Not what is interesting, but which branch would change the recommendation if it turned out wrong. Typically two or three branches carry the answer and the rest is cut.

5. It synthesises and pyramids. Answer first, three supporting reasons, evidence beneath, landing on a decision with an owner and a date.

6. It runs the quality gate on itself. A fact pass on every number and proper noun, a logic pass on the titles read alone, a client pass asking who in the room does not want this to be true and whether their best objection is answered, then a bias sweep. The linter runs on the file before you see it, not after you find the problem.

It matches effort to stakes. A question in chat gets two to six sentences, no framework and no headings. A board paper gets the full workflow and a red team pass.

Consultants and advisers who want the standard enforced on every draft, including their own.

Corporate strategy, transformation and M&A teams producing board papers, options papers and business cases without a firm behind them.

Private equity and investment professionals writing IC papers, value creation plans and portfolio reviews.

Founders, CFOs and operators who need to think through a question that matters the way a partner would, and who would rather be challenged than agreed with.

It also runs the everyday work. An email, meeting prep, the week ahead, a difficult conversation, a briefing note, deciding what to drop. Shorter register, same discipline.

A Claude account. Skills work on the Free, Pro, Max, Team and Enterprise plans, and in Claude Code. Code execution and file creation must be switched on under Settings, Capabilities.

No coding needed. Installation is uploading one zipped folder. The included guide shows each step.

Microsoft Office is optional. The agent works in chat on its own. Excel and PowerPoint are only needed when you want files produced, and the two review scripts read Word, PowerPoint, Excel and markdown.

On the build environment. The two review scripts run Python, so grading or producing files needs an environment that can execute it. Claude's own code execution environment is the tested one. Agent Skills is an open format, so the package is also read by ChatGPT and Codex, Gemini CLI, GitHub Copilot, VS Code and Cursor.

Single user licence. One named person, unlimited projects.

Commercial use included. Use it on client engagements and bill for the result. The analyses, models, reports and decks you produce are yours.

Team licences available. Write to info@wangadvisory.com for multi seat pricing.

No resale or redistribution. The package, its methods, templates and scripts may not be shared, published, resold, or used to build a competing product.

Professional responsibility stays with you. Every output is a draft working document for professional review. It carries no audit opinion and no third party assurance, and must not be presented as one.

Available soon. This product is listed ahead of release. Write to info@wangadvisory.com and you will be told the day it ships.

Instant download at release. The link appears on the confirmation page and arrives by email the moment payment clears.

Updates included. Improvements to a version you have bought are sent to the email address on the order at no extra cost.

Something wrong with the file? A failed download, a missing file or an incomplete package is replaced or refunded without an argument. Write to info@wangadvisory.com with your order number.

Customisation is a separate engagement. Your own house standard, templates, definitions, workstream structure, reporting format or sector calibration: these are scoped and quoted separately. Write to proposal@wangadvisory.com.