/start

description: Build your growth plan — researches your market and builds 7–9 unconventional tactics (~1.5h, free). Type /diffmode-growth-tactics:start your-site.com

Start — build your growth plan

Main-thread orchestrator for the **Diffmode free growth-ideation pipeline** — from a fast founder intake all the way to a final `synthesis.md` of **7-9 novel demand-gen tactic IDEAS**, and it **STOPS there**. (A best-effort packaging stage then re-writes `synthesis.md` as the founder-facing `growth-tactics.md` — same tactics, plain language, no new content.) Prioritization, implementation guides, and the proprietary 576-vector database are the paid product; this command never attempts them.

It runs the workflow as skills + worker sub-agents. It owns the DAG, the parallel fan-out, the reviewer-retry quality gate, and the stage-boundary checks.

> **This command runs in the main thread.** It dispatches worker sub-agents via the > Agent/Task tool, and it is the only place that may talk to the human (via AskUserQuestion, > for intake). It must NOT itself be run as a sub-agent — sub-agents are one level deep and > could not then spawn the workers. (See the repo's `docs/architecture.md` — internal design > notes, not shipped in the plugin.)

> **Naming under the plugin.** Ships in the `diffmode-growth-tactics` plugin; invoked as > `/diffmode-growth-tactics:start`. Use these exact plugin-namespaced ids in > Agent-tool dispatches: > - skills: `diffmode-growth-tactics:diagnostics-intake`, > `diffmode-growth-tactics:enrichment-<dimension>`, > `diffmode-growth-tactics:competitor-gaps` · `:cross-industry` · `:platform-arbitrage`, > `diffmode-growth-tactics:growth-factors-mining`, `diffmode-growth-tactics:lite-constraints`, > `diffmode-growth-tactics:synthesis-explore` · `:synthesis-build`, > `diffmode-growth-tactics:founder-report` > - workers: `diffmode-growth-tactics:research-worker`, > `diffmode-growth-tactics:analysis-worker`, `diffmode-growth-tactics:synthesis-worker`, > `diffmode-growth-tactics:reviewer` > The plugin is **self-contained**: it reads its bundled channel menu from > `${CLAUDE_PLUGIN_ROOT}/reference/` and writes all run outputs to a `./<slug>/` workspace in > the user's current directory. No host repo is required.

Step 0a — Welcome (print BEFORE any tool call)

The very FIRST thing this command does — before pre-flight, before any Bash/Read/Agent call — is print this welcome so the founder knows what's about to happen, how long it takes, and what they get back. Print it verbatim. **Markdown only** — headers, bold, lists, `---` rules; NO ASCII art, NO ANSI codes (rendering is terminal-dependent; markdown renders reliably).

> # Diffmode — let's build your growth plan > > Here's what happens next: > > 1. **A few quick questions** (~2 minutes) — the things your website can't tell us. > 2. **Hands-off research** (~60–90 minutes) — we study your competitors, your buyers, > and what's already working in your market. You can walk away. > 3. **Your growth tactics** — 7–9 specific ways to get users, built for your budget, > team, and stage. > > **What you'll have at the end:** > > - **Your Growth Tactics** — the main event: each tactic with first steps and an > early signal to watch. > - **3 research briefs** — Competitor Research, an Audience Map, and an Acquisition > Audit. Reusable on their own. > - **3 strategy reports** — where your size wins, plays from other industries, and > fresh platform openings. > - **Working papers** — the notes behind the work, yours to keep. > > These won't be the tactics you'd get by asking an AI for growth ideas — anything an > average marketer would suggest gets thrown out along the way. > > **Total: usually 1–1.5 hours.** You only need to be here for the questions at the > start; results open in your browser at the end.

(Fill in nothing; the block is static, and it always prints in full — auto-resume detection needs tool calls, so it happens after (Pre-flight step 1). When the founder picks **Continue** there, follow the choice with one short line: what's already done + the time estimate for the remaining stages.)

User-facing voice (applies to EVERYTHING you print during the run)

The founder is a marketer or a busy founder, not a developer. The routines below define what to DO; this section defines what to SAY. When they differ about what to print, this section wins.

• **Stage start = ONE plain line with an ETA.** Stage done = ONE line with a human

metric. Suggested lines (vary naturally, keep the shape):

| Stage | Start line | Done line | |-------|-----------|-----------| | diagnostics | "Looking at your site and getting set up (~3 min)…" | "✓ Got your product brief" | | enrichment competitors | "🔍 Researching your competitors (~15–20 min)…" | "✓ Competitor research done — 9 competitors mapped" | | enrichment wave 2 | "🔍 Mapping your buyers + auditing what works in your market (~15–20 min)…" | "✓ Buyer map + acquisition audit done" | | growth-factors | "⛏️ Mining growth mechanisms from public case studies (runs in the background, ~40 min)…" | "✓ 27 growth mechanisms mined" | | think-tanks | "🔭 Working three strategy angles in parallel (~15–25 min)…" | "✓ 3 strategy reports done" | | lite-constraints | "Setting up the tactic builder (~3 min)…" | (fold into the next start line) | | synthesis | "🧪 Building your tactics (~25–35 min)…" | "✓ 8 tactics built — passed the quality check" | | founder-report | "📦 Packaging your report (~3 min)…" | "✓ Your Growth Tactics ready to read" |

• **NEVER narrate to the founder:** structural-check internals, run-ledger writes or

any JSON, reviewer scores/verdicts/rubric mechanics, vector counts-as-plumbing (`must_include`, pools, anchors, IDs), retry/respawn logistics, worker lifecycle, or model tiering. All of that detail still goes into `WS/.run-state.json` exactly as the routines specify — the ledger is unchanged; only the narration is quiet.