preview

Competitive Research — Preview

Use this when the user wants to sanity-check what the full analysis will base its work on — the inferred product identity and the competitor list — before running the full Phase 3–8 deep dive. The full run is 10–20 minutes; this preview is 1–2 minutes and prevents wasting time on a wrong product or competitor model.

The dossier and competitor list this skill writes are reused by the main analysis skill — running the full analysis afterwards picks up where preview left off rather than redoing Phase 0 and Phase 2.

Workflow

1. Check for an existing dossier from today

If ./competitive-research/runs/YYYY-MM-DD/product-dossier.md exists for today's date, read it and ./competitive-research/competitors.yaml instead of regenerating. Skip to step 4.

If they exist but the user wants a fresh look, ask first — don't silently overwrite.

2. Run Phase 0 (product dossier)

Execute the same Phase 0 logic as the main analysis skill. The detailed playbook is at ${CLAUDE_PLUGIN_ROOT}/skills/competitive-research/references/codebase-discovery.md — read it and follow it to infer:

• Product name, one-liner, URL, target user, pricing model
• Core JTBDs (2–4 in Christensen form)
• Canonical feature list
• Inferred strategic constraints (with evidence sources)
• In-repo competitor mentions

If ./competitive-research/overrides.yaml exists, respect those overrides instead of inferring those fields.

Write the dossier to ./competitive-research/runs/YYYY-MM-DD/product-dossier.md using the format from the codebase-discovery reference (section 0.7).

3. Run Phase 2 (expand competitor set)

Same as Phase 2 of the main analysis skill (read ${CLAUDE_PLUGIN_ROOT}/skills/competitive-research/SKILL.md Phase 2 for the full instructions). Aim for 8–15 competitors covering Direct / Adjacent / Indirect / Emerging categories. Save to ./competitive-research/competitors.yaml.

4. Print the preview

Show the user, in this order:

1. Product summary — name, one-liner, target user, top 2–3 JTBDs (one line each)
2. Strategic constraints — the 3–6 inferred bullets
3. Competitor list — full set with one-liners, grouped by category
4. Confidence flags — any inferences the dossier marked as low-confidence (target user often is)

Keep the print compact. Don't reproduce the entire dossier — point at the file path for the full version.

5. Ask the user to confirm or correct

End with this exact prompt:

Does this look right?

• Yes, run the full analysis — say "run the competitive research" and the main skill will pick up the dossier + competitors from here
• Some inferences are wrong — run /competitive-research:setup to set overrides, then preview again
• Add/remove specific competitors — tell me which, and I'll update competitors.yaml directly

Do not proceed past Phase 2 from this skill. The full analysis is a separate invocation by design.

Quality bar

• Phase 0 outputs must be grounded in actual file reads, same standard as the main skill.
• Don't fabricate a competitor list — every entry needs a URL.
• Don't run any phase past Phase 2. Stop and ask.
• If the user has an overrides.yaml, respect it. Don't re-infer fields they've already set.