validate-competitors

"Is this already solved?" is the wrong question. Something always exists. The right question is what the people who use it still complain about. This skill mines each competitor's own users with the same tools and the same codebook as your main run, so the two line up theme by theme.

What the user typed after /validate-competitors: $ARGUMENTS

The tools

python3 "${CLAUDE_SKILL_DIR}/gapmap.py" [--dir validator]
python3 "${CLAUDE_SKILL_DIR}/../validate-mine/sources.py" --dir <folder> <command>
python3 "${CLAUDE_SKILL_DIR}/../validate-cluster/batches.py" --dir <folder> <command>

If the path is not expanded, use the base directory Claude Code printed for this skill. Below, MINE, BATCHES and GAPMAP mean these three.

Step 1: pick 2 to 4 competitors

Start from what the data names: search.competitor_apps in idea.json, the tried lists in themes.json, and tools named in wtp.md. Add any the user names. Ask the user to confirm the list. Each one costs a mining run and one labelling sub-agent per 50 complaints.

Step 2: the facts, each with a source

For each competitor, find what it does, who it is for, its price and its platforms. Use web search and the competitor's own pages. Every fact gets the link it came from and today's date. If you cannot find a fact, write "unknown". Never fill it in from memory: prices change, and a wrong price in a report is worse than a blank.

Step 3: mine each one

Use a short lowercase slug (beeper). For each competitor:

mkdir -p validator/competitors/<slug>
cp validator/idea.json validator/cluster_codebook.json validator/competitors/<slug>/
MINE --dir validator/competitors/<slug> appstore --app <id> --name '<Name>'
MINE --dir validator/competitors/<slug> reddit --query '"<Name>"' --since <SINCE> --max 300
MINE --dir validator/competitors/<slug> hn --query '"<Name>"' --since <SINCE> --max 200

Use find-app to get the App Store id. Skip sources that do not apply. The same rules as /validate-mine hold: run the calls one at a time, keep partial runs, and never type posts in by hand.

Step 4: label each one with the same codebook

BATCHES --dir validator/competitors/<slug> plan --batch 50

Run the waves exactly as in /validate-cluster, then check and tally with the same --dir. Never edit the codebook copy in a competitor folder. The gap map refuses a competitor labelled with a different one, because the themes would not line up. If the main codebook changes, copy it again and relabel with --fresh.

Step 5: the gap map

GAPMAP

It writes validator/gaps.md and gaps.json. For each of your themes it shows your pain score and the share of each competitor's complaints about it, then a status:

• open gap: you scored it as real, and at least a tenth of a competitor's complaints are about it. Their users still hurt here. This is where to win.
• they struggle too: their users complain, but your data did not score it as real.
• maybe solved: under 3% of every competitor's complaints, with enough data.
• unclear: anything else, including a competitor with fewer than 20 complaints.

It also lists what their users complain about that your codebook does not cover. Read that list. A new theme there can be a gap you have not seen yet.

Step 6: write validator/competitors.md

1. A facts table: name, what it does, who for, price, link, date checked.
2. The gap map summary: open gaps first, each with one quote from gaps.md.
3. What their users raise outside your codebook.
4. A one-line position: "For <beachhead> who <top pain>, unlike <competitor>, whose users say <quoted complaint>."

Quote only from gaps.md. Those quotes passed a word-for-word check.

Honest limits, say them once

Shares are of complaints, not of users: a happy majority writes nothing. App Store reviews come from one country and default to 3 stars and below. A competitor with few complaints may be small rather than good. "Maybe solved" means quiet, not proven.

Output

validator/competitors/<slug>/ for each competitor, validator/gaps.md, validator/gaps.json, validator/competitors.md.

Next: /validate-prospects finds the people to interview.