market-research

When to invoke this skill

Answers a market question in service of one decision: "should we launch in Dhaka", "is there demand for a cheaper tier", "who buys this and what do they pay". Use when asked to "research this market", "is there a market for…", "should we enter…". Umbra's skill. For a standing comparison of named rivals, use /competitor-watch.

The decision comes first — the hard boundary

Research without a decision is a pile of links. Before reading a single source, agree the decision in one sentence ("Decide whether to offer a Bangla-language plan by December") and what would change it (two or three things that, if true, swing it). Write both at the top of the file. Every source you read earns its place by bearing on one of them.

• True vs guessed, always visible. A fact carries the URL it was read at and the date read. An estimate is labelled guess: with the reasoning that produced it. A number you did not read on a page — market size, growth rate, share — is never written as if you had.
• UNKNOWN is a finding. What you looked for and could not find is listed, with where you looked.
• No signing in, no paid reports. Gated reports and sites are read only through what public pages and search results say about them; a paywalled figure quoted second-hand is cited to the page that quoted it, labelled so.

What you need first

• The decision, in one sentence, and the deadline if there is one.
• What the business sells, in their words — DESIGN.md or a PLAN-*.md may already say.
• Scope — the place, the customer, the time frame.

Ask for everything still missing in a SINGLE message, then wait. Without a decision, propose one from the request and confirm it before researching.

Procedure

1. Frame. Write the decision, what would change it, and the questions that follow (at most six). Pick the slug: <topic> from the decision (bangla-plan).

Scratch for this run lives in .localstack/work/{date}-{slug}/ — hidden, one directory per run, so a new run never clobbers an earlier one and the folder's top level stays the standard files. Scratch is disposable; old run directories may be deleted freely.

2. Read. For each question: search, then fetch the pages that actually answer it — official statistics, the companies' own pages, reviews and forums for what buyers say, job posts and news for direction. Note each source in .localstack/work/{date}-<topic>/sources.md (URL, date read, what it says, which question). Prefer primary sources; a blog repeating a number is cited, but the number is traced to where it began when you can.

3. Sort every claim into True (read, cited), Guessed (inferred — say from what) or Unknown (looked, not found — say where).

4. Write research-<topic>.md at the top of the working folder:

# Research — <topic>
Updated: YYYY-MM-DD

## The decision
<one sentence> — what would change it: <2–3 things>

## What's true
- <claim> — <URL> (read YYYY-MM-DD)

## What's guessed
- guess: <estimate> — because <reasoning, citing the true items it rests on>

## Unknown
- <what> — looked at: <where>

## What it means for the decision
<3–6 short points, each pointing back at the true items it rests on, and
which way it pushes the decision; then the one thing to find out next>

A rerun replaces the file; CHANGELOG.md and git keep the history.

Standard files. This folder is kept in files any agent already reads. Update them in place; never scatter output into new folders.

• AGENTS.md — create it if missing. localstack owns only the block between <!-- localstack:start --> and <!-- localstack:end -->; rewrite that block, never anything outside it. The block says what this folder is for, the rules (drafts only; nothing is sent without the user's explicit yes, one message at a time; no invented facts), a map of the files below, and one line per topic (its PLAN, its lead count, the next unticked step) and per report (its file and date).
• CHANGELOG.md — create it if missing (# Changelog). Add one bullet for this run under today's ## YYYY-MM-DD heading, newest date first: the skill, the topic, and the counts or outcome (e.g. - lead-search austin-dentists: 18 found, 3 skipped as already contacted).
• TODOS.md — create it if missing (# TODOs). Add each open next action as - [ ] <action> (<topic>); tick items this run finished; never delete lines.
• DESIGN.md — decisions meant to last (positioning, tone, channels to use or avoid). Read it before writing anything a person will see; add to it only when the user states or approves a decision.

For this step: the CHANGELOG line gives the decision, sources read, and true / guessed / unknown counts. One TODO per unknown that would change the decision (- [ ] find out: <unknown> (research <topic>)). Record the decision in DESIGN.md only when the person makes it.

5. Report. Lead with what it means for the decision, then the biggest unknown. Offer /competitor-watch when named rivals came up.

Completion status. End the chat report with one of:

• DONE — completed, with the evidence named (files written, counts, URLs).
• DONE_WITH_CONCERNS — completed, and list each concern.
• BLOCKED — cannot proceed; say what blocked it and what was tried.
• NEEDS_CONTEXT — missing information; say exactly what is needed.

Quality bar

• Every line under "What's true" opens to a page that says it.
• No number appears that was not read on a cited page or labelled a guess with its working.
• The implications answer the decision asked — not a general market essay.