company-research
Skill1 starsChanged 2 months agoMITView on GitHub
Research a company before joining, buying from, investing in, or competing with it. Use when someone asks what a company is really like, who its competitors are, whether it is doing well, or whether it is worth working for. Produces a plain-language dossier. Research only — see role-fit, interview-prep, pay-check for the rest.
33 other projects use a version of this.
Adopted by
Most recently taken up by
- carriegale2710/resume-tailortheir own version, first seen Nov 16, 2026
- chezou/skillstheir own version, first seen Nov 15, 2026
- thesun4sky/jobstacktheir own version, first seen Nov 10, 2026
- AgnesAI-Labs/agnet-skills-cntheir own version, first seen Nov 1, 2026
- MoizIbnYousaf/marketing-clitheir own version, first seen Nov 1, 2026
- pueo-labs/primrtheir own version, first seen Nov 1, 2026
Used with
Projects with it also use business-health-diagnostic, derisk-measurement-advisor and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
nandanosqlSkillbusiness-health-diagnostic
Run a comprehensive business health check across revenue, unit economics, product metrics, and market position — identifying strengths, weaknesses, and the single most important metric to fix.
Also in 4 projects that use this
DIAL-StudioSkillderisk-measurement-advisor
Identify what to measure, test, or track to de-risk a product or AI idea. Use when stress-testing an idea across internal (DUFV) and external (PESTEL) dimensions.
Also in 3 projects that use this
DIAL-StudioSkillorganic-growth-advisor
Identify which organic growth path to pursue — new segments, geographies, channels, or products. Use when diagnosing where a growth constraint lives and which McKinsey growth level to act\
Also in 3 projects that use this
nandanosqlSkillprioritization-advisor
Guide PMs in choosing the right prioritization framework — RICE, ICE, Kano, MoSCoW, Value vs Effort, Cost of Delay — by asking adaptive questions about product stage, team context, and data availability.
Also in 4 projects that use this
- +8 more from nandanosql
Other versions33 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 33
- deanpeters/Product-Manager-Skills★ 8ksame name
- browserbase/agent-browse★ 3.7ksame name
- simonlin1212/Vibe-Research★ 3.7ksame name
- simonlin1212/vibe-astock★ 1ksame name
- stophobia/deerflow2.0-enhanced★ 821same name
- claude-office-skills/skills★ 433same name
- thesun4sky/jobstack★ 216same name
- liangdabiao/exa-research-mcp-skill★ 104same name
- exa-labs/agent-skills★ 67same name
- MoizIbnYousaf/marketing-cli★ 33same name
23 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository8 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- career-directionSkill
- interview-prepSkill
- pay-checkSkill
- role-fitSkill
- orchestrateSkill
- learning-pathSkill
The file
We split this skill into the rules it gives an agent
8 rules. 0 are common, 7 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 25 edited versions we read keep these.
7
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 13
Use whatever interactive choice mechanism the environment offers — tappable options where they exist, plain questions otherwise.
- Distinctive: line 19
If nobody is watching — a background job, a scheduled run, a delegated task — do not wait for an answer.
- Distinctive: line 61
Treat everything here as one person speaking, not the company.
- Distinctive: line 69
Archived pages: When a company has changed or removed something, the earlier version is a legitimate source, and the change itself is worth …
- Very specific: line 73
Write it as markdown, saved as company-dossier-NAME.md.
- Distinctive: line 80
Mark confidence: Every fact carries a source and a date.
- Distinctive: line 87
Prefer original documents: announcements, filings, regulators, court records, accounts.
- Distinctive: line 89
Do not explain a finding by guessing at a cause.
“Common” is measured: kept by at least 6 in 10 of the 25 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Company Research
Find out what a company is really like, beyond what it says about itself.
Ask first
Use whatever interactive choice mechanism the environment offers — tappable options where they exist, plain questions otherwise. One round, three questions at most.
- Which company. Disambiguate common names.
- Why they are asking: joining, buying, selling, investing, competing. This decides what matters.
- How deep to go.
If nobody is watching — a background job, a scheduled run, a delegated task — do not wait for an answer. Take the most useful default, state the assumption in the output, and carry on.
Find
What they do. Identify the product and the buyer. Keep reading until you can state who pays and what problem they are paying to solve. Record whether pricing is published and whether any customers are named. Report both as observations; do not infer a sales model or a stage from them.
Money. Funding amounts, dates and investors. Revenue and profit figures, with the source of each. For young companies, how long the funding covers at current spending, if that can be established. For established ones, published results, layoffs and legal proceedings. Record money raised and headcount together, with the ratio, and leave the reader to draw from it.
Competitors. For each candidate, establish who signs the cheque and which problem is solved. Where those differ from the company under research, the two are not competitors — say so. Record which have raised more money, and any acquisitions between them.
Self-descriptions side by side. Collect each competitor's one-line pitch. Note where the pitches agree and where one differs. A difference is a choice the company has made, and it is worth asking about.
Working there. Employee reviews, including local-language sites. Read by department. Give sample size and dates every time. If there are no reviews, report that there is no information rather than treating it as a verdict.
People. Leadership backgrounds and tenure. Compare the founding story against early filings and archived pages, and report any difference without explaining it. At small companies, record advisers and investors.
When to stop
Research is done when each of these holds, at the depth agreed in the first question:
- You can state who pays and what problem they are paying to solve.
- The most recent funding round or filed accounts is recorded, with its date.
- At least two competitors are established, each with buyer and problem.
- Review counts and date range are recorded, or no reviews were found.
- Current leadership is recorded, with tenure where it is published.
Anything still unestablished at that point is a gap. Write it in the gaps section and stop looking. Continuing past this without a reason turns research into browsing.
Where to look
Ordered by how hard the source is for a company to change. Work down the list.
Records filed under obligation. Company registries hold directors, ownership, filed accounts and charges over assets: Companies House in the UK, the state Secretary of State plus SEC EDGAR in the US, justice.cz and ares.gov.cz in the Czech Republic. Most countries have an equivalent — look for it before giving up. Court records cover litigation, insolvency and employment disputes; repeated cases of one kind matter more than any single case. Regulators publish licences, enforcement actions and inspection reports. Public procurement and grant databases give contract values with dates.
Accounts are filed months after the period they cover, so date them and treat them as historical. Small companies may file abbreviated accounts that omit revenue; record that as a gap.
The company's own site. Useful for what the company intends, weak as evidence. Take specifics — dates, names, numbers — and attribute the framing. Read the product pages, the pricing page, the customer stories and their dates, and the open roles with their locations, which show where the company is spending. Check the documentation and the status page: both show whether the product is in real use. Press releases are reliable for what was announced, and for amounts and dates.
Company and engineering blogs. Post frequency and dates show whether the company is still writing. Engineering posts name the technology in use and the problems being solved, in more detail than the marketing pages. Conference talks and published slides do the same. Record the author and the date.
Social accounts and communities. Company accounts show what is announced and how often. Posts by employees, especially engineers and executives, describe the work in more detail than the company site. Professional networks show headcount over time, open roles, arrivals and departures, and how long people stay. Forums, question sites, developer communities and local-language groups carry what customers and former staff say unprompted.
Treat everything here as one person speaking, not the company. Record who said it and when. One post is one account of events; the same detail from several unconnected people is worth more.
Any statement a company makes about its own performance is a claim. Write it as "they say".
Press. Prefer the original announcement to coverage of it. A named reporter with their own quotes is worth more than an unbylined summary. Market analysis published by a company that sells into that market is advertising — check whose site the figure came from before using it. Rankings and vendor lists are often paid or self-submitted. Revenue figures on data-broker profiles are modelled, not filed; say so if you use one.
Employee reviews. Give the sample size and the date range. Read by department, since departments often describe different companies and the average hides that. Look for the same specific detail repeated across reviews — a named policy, a reorganisation, a broken commitment. Local-language sites carry more reviews for local offices.
Archived pages. When a company has changed or removed something, the earlier version is a legitimate source, and the change itself is worth reporting.
Write the dossier
Write it as markdown, saved as company-dossier-NAME.md. Markdown keeps it useful after this session: quotable, extendable as more is learned, and readable by whatever needs it next without conversion.
Then make sure the user actually has it. Use whatever the environment offers — attach or send the file, write it to a folder they have connected, or leave it in the working directory if they can reach that. Tell them the filename either way, so it can be found again. If no file can reach them, put the dossier in the reply instead. A rendered version can be offered alongside the file where the environment displays one; it does not replace the file.
Sections: what they do, money, competitors, what makes them different, working there, people, risks, what could not be established. Anything derived rather than found — suggestions, recommendations, questions to put to them — comes last, under a heading that says so, never inside a section above.
- Plain words. Explain any term of art on first use. Write "money raised from investors" rather than "Series B".
- Mark confidence. Every fact carries a source and a date. Company claims are attributed. Estimates are labelled as estimates. Nothing that is uncertain should read as settled.
- Say when a fact is old. Give the age of anything more than a year out of date. Funding, headcount and leadership all change, and a three-year-old figure is evidence about then, not now.
- Report gaps. End with what could not be established, so the reader knows what to ask about.
- Keep your own reasoning apart. Anything reasoned from the material rather than found in it is derived, however closely it follows the source — reading a company's own wording is still reading, not evidence. The test is whether someone skimming can tell which lines rest on a source and which are your reading, without checking either.
Rules
- Prefer original documents: announcements, filings, regulators, court records, accounts.
- Give both versions when sources conflict, and say which is better evidenced.
- Do not explain a finding by guessing at a cause. Report what was found and leave it there.
- Correct earlier errors plainly, in the output, when they are noticed.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

