I stress-tested an AI equity research skill on ClawhHub with a question that stumps most analysts

The question I had in mind before I even opened the page: How do you forensically evaluate earnings quality for a high-SBC software company where GAAP net income is deeply negative but management keeps citing adjusted EBITDA?

That's not a beginner prompt. It's something that trips up even experienced investors who know the financials but can't quite articulate the manipulation vectors. I wanted to see if a skill-packaged AI terminal could surface real depth -- or just regurgitate a textbook paragraph.

Installing the Drillr skill from ClawhHub

I went to clawhub.ai/yx9966/drillr, starred it, and hit download. The whole friction from landing on the page to having the skill active was under two minutes. ClawhHub gives you the install command inline -- no hunting through docs.

# ClawhHub skill install (from the page)
clawhub install yx9966/drillr

After that, the skill is invokable directly. Drillr exposes a clean interface: you pass a natural language question, it POSTs to a hardcoded Google Cloud Run endpoint, and streams back markdown. No setup beyond the install. No API keys to wrangle. No config files.

The skill's SKILL.md was honest about what it does: it's a read-only financial research terminal for US public equities. Thesis-driven screening, forensic accounting, SEC filing extraction, earnings call analysis. The scope is clear, which I appreciated. No pretending it can do things it can't.

The question I chose and why

I didn't want to ask something like "what is the P/E ratio of Apple" -- any LLM wrapper can handle that. I wanted to probe the forensic accounting angle specifically, because that's where most tools collapse into generic advice.

My question:

"Walk me through how to identify earnings quality deterioration in a software company that reports negative GAAP net income but positive adjusted EBITDA. Specifically, what line items in the 10-K should I reconcile, and what patterns in stock-based compensation and deferred revenue changes would signal that adjusted metrics are masking real cash generation problems?"

This is a real question. I've been thinking about it in the context of a few small-cap SaaS names where the EBITDA story sounds compelling until you actually pull the cash flow statement.

What came back

The response was structured as a multi-part breakdown: first, a reconciliation walkthrough from GAAP net income to operating cash flow (not EBITDA -- it correctly noted that EBITDA excludes capex and working capital movements that matter enormously in SaaS), then a specific list of SBC-related red flags to look for.

The SBC section was the best part. It called out the pattern where SBC as a percentage of revenue increases even as the company claims to be scaling -- a sign that dilution is accelerating, not moderating. It also flagged deferred revenue drawdown as a signal: if a company is pulling forward recognized revenue while new bookings slow, the adjusted EBITDA number can look stable while the business is quietly deteriorating.

It didn't just dump text. It told me what to look for and where to find it in the filing. That's a different kind of answer.

Time to answer

From hitting enter to a complete streamed response: roughly 14 seconds.

For a question of that complexity, that's fast. I've had slower responses from ChatGPT on simpler prompts. The streaming format helped -- I was reading before it finished generating.

Am I satisfied?

Yes, with one caveat.

The answer was genuinely useful. It covered ground I'd partially thought through and added two angles I hadn't explicitly framed -- specifically the deferred revenue velocity point and a note about capitalized software development costs as another GAAP/non-GAAP wedge. That's the kind of thing you find in a well-annotated equity research report, not a generic AI response.

The caveat: the skill routes your queries to an external endpoint. The SKILL.md is upfront that you should treat query content as potentially logged. For anyone doing proprietary research on undisclosed positions, that's worth knowing. For general learning and public company analysis, it's fine.

Who should actually use this

If you're a developer who also manages your own portfolio -- or you're building tooling for an investment workflow and need to understand the research domain better -- this is worth 90 seconds to install and try. It's not a Bloomberg Terminal. But as a question-answering layer on top of public equity data and filings logic, it punches above what I expected from a ClawhHub skill.

The install is clean, the scope is honest, and it answered a hard question well. That's enough to keep it in my toolkit.

Tested on ClawhHub via clawhub.ai/yx9966/drillr. Skill by Drillr.ai.