mitchell-agentic-sprint

🧢 Mitchell Agentic Sprint

A Claude Code plugin that walks an AI-strong builder through six sequential steps to take one of their many ideas and prove out whether it can become a real business — adversarial by default, refusing flattery, demanding verbatim evidence.

By the end of the Sprint, the builder has a locked Vertical Stack, 15–30 buyer interviews captured verbatim, a Mom Test–disciplined theme map, an adapted expert framework, a competitor cluster map with positioning whitespace, and three markdown artefacts ready to take into customer conversations and pre-seed pitches: a sales deck, a customer outreach plan, and a pre-seed investor deck.

The Sprint does not build the product or the website. It gets the builder to the point where they know whether what they're building will sell, and they have the materials to prove it.

Concept

Authored by Scott Mitchell (Mitchell Agentic Partners). Working name.

→ Full concept: docs/concept/concept.md

Status

v0.3.2 — Claude Code + Cowork plugin testing. Ironed out a few bugs, trying out the workflow to see if it works end to end!

Design principles

• Adversarial by default — every prompt refuses flattery and pushes back on vague answers (concept §6)
• Behavioural over hypothetical — interview scripts pass through a Mom Test linter that bans "would you…" questions before the user ever sees them (SPEC-008-A)
• Saturation over headcount — interviews stop when the last 3 surface no new themes, not at an arbitrary count (helpers/saturation.py)
• Loopback on validation failure — failed verticals retain Builder Profile, themes, and framework into the next attempt; Vertical Stack and downstream are cleared (concept §7, SPEC-005-A)
• Mini Council — five skeptical NPC voices (Seed VC, Churned Customer, Competitor Founder, Future-Self, Grumpy PM) tear up the idea between Steps 1→2 and 4→5
• Output is artefacts, not product — three markdown artefacts populated from sprint state, no padding when evidence is missing (ADR-009)

Quick start

Claude Cowork (recommended for MAS)

The Sprint is well-suited to Cowork. A six-step validation pipeline runs across multiple sessions over days or weeks, with persistent state at each gate and asynchronous artefact production — that's the shape Cowork is built for. Use Cowork unless you have a reason to prefer the CLI.

1. From the Cowork tab in Claude Desktop, select + → Plugins → Add plugin

2. Add marketplace — Select the By Anthropic dropdown, then Add marketplace from GitHub and enter:

   https://github.com/cgbarlow/skills
   

3. Install three plugins from the marketplace: Six Animals, Campaign Mode, and Mitchell Agentic Sprint (the Sprint depends on the other two; per ADR-003)

4. Begin your Sprint — run:

   /sprint-start
   

   On the first run in a project, /sprint-start auto-invokes setup: prerequisites are verified and the 🧢 Mitchell profile is offered (default install). The Sprint runs in Claude-only synthesis mode — fully featured for Cowork users. (The optional NotebookLM bolt-on for source-grounded synthesis is Claude Code CLI only; see below for that path.) Then Step 1 (Discovery) opens with Mitchell framing your unfair advantages, runway, and rough idea.

   To re-configure later (install Mitchell after declining, onboard NotebookLM after the fact), call /sprint-setup directly.

Claude Code CLI

For terminal-first users:

1. Install Claude Code (full guide):

   curl -fsSL https://claude.ai/install.sh | bash
   

2. Add the marketplace — in a Claude Code session:

   /plugin marketplace add cgbarlow/skills
   

3. Install the three plugins:

   /plugin install six-animals@cgbarlow-skills
   /plugin install campaign-mode@cgbarlow-skills
   /plugin install mitchell-agentic-sprint@cgbarlow-skills
   /reload-plugins
   

4. Begin your Sprint:

   /sprint-start
   

   First run auto-invokes setup (prerequisites, Mitchell profile, NotebookLM offer). Re-configure later via /sprint-setup.

Pinned compatibility: campaign-mode@0.4.8, six-animals@0.1.2 (verified 2026-05-08). Or clone the repos into your .claude/plugins/ directory if you'd rather skip the marketplace entirely.

Optional (Claude Code CLI only): NotebookLM MCP bolt-on