Etudes — Sprint Coach Plugin

A sprint coach for builders who have more ideas than shipped products. Etudes interviews you about your project, skills, and patterns, then generates a 5-day sprint calibrated to how you actually work.

Overview

Etudes provides a structured coaching workflow inside Claude Code. It reads your codebase, conducts a conversational intake interview, detects avoidance patterns, and generates a concrete 5-day sprint with time-estimated tasks grounded in your actual files. During the sprint, it provides daily check-ins, catches avoidance in real-time, and captures new ideas in a parking lot so they don't derail focus.

All state persists in a .etudes/ directory at your project root, so coaching continues across sessions.

Contents

| Type | Name | Description | |------|------|-------------| | Skill | etudes | Core coaching engine — intake interview, pattern detection, sprint generation, active coaching | | Command | /etudes | Main entry point — runs intake if new, shows sprint status if returning | | Command | /etudes-checkin | Daily check-in — reads sprint, asks what's done, updates progress | | Command | /etudes-retro | Sprint retrospective — reviews what shipped, parking lot, generates next sprint | | Command | /etudes-park | Parking lot capture — saves an idea mid-sprint without context-switching | | Command | /etudes-dashboard | Cross-project status view in Claude Code (reads all registered projects) | | Local App | etudes-dashboard | Web dashboard at localhost:2400 with kanban + park/icebox across projects |

Commands

/etudes

Launches the full workflow. On first run, conducts the intake interview and generates a sprint. On subsequent runs, reads .etudes/ state and shows current sprint status.

Usage:

/etudes

Or with context:

/etudes I'm building a recipe app and I keep rewriting the data model instead of shipping

/etudes-checkin

Daily check-in during an active sprint. Reads the current sprint file, determines what day you're on, and asks for a status update. Marks completed tasks and redirects if it detects avoidance patterns.

Usage:

/etudes-checkin

Or with an update:

/etudes-checkin finished tasks 1 and 2, stuck on the auth flow

/etudes-park

Captures a new idea mid-sprint without derailing focus. Appends to .etudes/parking-lot.md and immediately redirects back to the current task.

Usage:

/etudes-park what if I added a dark mode toggle

/etudes-retro

End-of-sprint retrospective. Cross-references completed tasks with git commits, reviews the parking lot, identifies patterns, and optionally generates the next sprint with adjustments.

Usage:

/etudes-retro

/etudes-dashboard

Cross-project status command inside Claude Code. Reads ~/.etudes/projects.json, aggregates all active projects, and shows progress + next tasks in one view.

Usage:

/etudes-dashboard

Project-specific drill-down:

/etudes-dashboard liminal

Local Dashboard (Option C)

Etudes now includes a local web dashboard that tracks all registered projects.

Features

• UI primitives powered by Basecoat (basecoat-css) with Motion-enhanced transitions
• URL routing:
  • http://localhost:2400/<project-slug>
  • http://localhost:2400/all
• 5-column kanban: Backlog, To Do, In Progress, Done, Shipped
• Task CRUD: create, edit, move status, comment, delete
• Park/Icebox panel across projects
• Direct file mutation of .etudes/sprint-current.md and .etudes/parking-lot.md
• Deleted tasks are caught by Etudes check-ins ("done, descoped, or avoided?")

Run locally from this repo

npm install
npm start

Then open:

http://localhost:2400

Publish for npx etudes-dashboard

This is a separate npm package step from npx skills add.

# one-time npm login
npm login

# publish package (from repo root)
npm publish

After publishing:

npx etudes-dashboard

The Workflow

Phase 1: Intake Interview

Etudes asks ~8 questions, one at a time, conversationally. It simultaneously scans the repo for signals: git log activity, README, tech stack, test coverage, deployment configs.

Entry path detection:

• Existing code: Assesses project state, asks where you get blocked
• Spec or PRD: Probes for the MVP buried in the spec
• Idea only: Asks what's stopped it from happening
• Multiple projects: Runs a lightweight dialectic to pick the shortest path to something shipped

Builder profile questions:

• What does "done" look like in 7 days?
• Technical background
• Work patterns (multi-select avoidance pattern detector)
• Shipping history
• Available time per day
• Preferred coaching tone

Phase 2: Pattern Detection

Based on intake signals, Etudes selects a coaching mode internally. It never announces the mode — it just shifts behavior.

| Mode | Triggers | Behavior | |------|----------|----------| | Architect to Executor | Elaborate plans, nothing shipped, overscoped goals | Cut scope aggressively. Trivially small first tasks. No spec editing rule. | | Confidence Builder | Self-taught, minimizing language, discounts shipped work | Validate with evidence from their code. Progressive difficulty. | | Focus Lock | Multiple projects, new ideas mid-conversation | Name the pattern. Redirect every time. Parking lot everything. | | Unblocking | Stuck on specific task, emotional language about blocker | Break into 10-minute chunks. Remove decisions. Reference specific files. | | Accountability | Git log gaps, vague about activity, shame language | Pick up where they left off. No shame. No lectures. |

Modes shift mid-sprint based on observed behavior.

Phase 3: Sprint Generation

Generates a 5-day sprint with tasks grounded in the actual codebase.

Sprint structure:

• Day 1 is always the easiest (build momentum)
• Each task has a time estimate and "done =" definition
• Tasks reference actual files when a repo exists
• Day 5 is always "Ship Day" — put something visible in front of a real person

Task calibration:

| Signal | Rule | |--------|------| | 30 min/day | 2 tasks, ≤15 min each | | 1 hour/day | 3 tasks | | 2-3 hours/day | 4-5 tasks | | "I get overwhelmed" | First task < 10 min, daily warm-up | | "I pivot to re-planning" | Sprint rule: no spec editing | | "I get pulled to new ideas" | Sprint rule: use /etudes-park | | Time varies wildly | Starred must-do task + optional full-day |

Sprint 1 is always labeled "Calibration Sprint" — the first sprint is about learning how you work.

Phase 4: Active Coaching

When the user returns for check-ins, Etudes responds based on context:

| Situation | Response | |-----------|----------| | Completed tasks | Mark done. "What's next?" | | Partial completion | "Which ones? What's blocking the rest?" | | Gap (missed days) | "What's left on Day [X]?" No shame. | | New idea mid-sprint | Park it. Redirect to current task. | | Re-planning detected | "This is the pattern. Next checkbox?" | | Frustration/anxiety | Zoom to smallest task. "10 minutes. Go." | | Wants to quit | "What specifically isn't working? Fix the sprint, not abandon it." |

Phase 5: Sprint Retro

Reviews completed vs incomplete tasks, cross-references git commits, walks through the parking lot, and generates the next sprint with adjustments if requested.

State Persistence

All state lives in .etudes/ at the project root:

.etudes/
├── profile.md           # Builder profile (coaching mode, tone, rules)
├── sprint-current.md    # Active sprint with checkboxes
├── parking-lot.md       # Ideas captured mid-sprint
└── retros/
    └── sprint-1.md      # Sprint retrospectives

This persists across Claude Code sessions. The commands read state on every invocation.

Global cross-project index lives at:

~/.etudes/projects.json

Dashboard-only kanban metadata is stored per project at:

.etudes/board-state.json

When to Use This Plugin

Use for:

• Side projects that have stalled
• Projects where you keep planning instead of building
• When you're torn between multiple projects
• When you need external accountability to ship
• When you know what to build but lose momentum in execution

Don't use for:

• Active team projects with existing sprint processes
• Quick bug fixes or one-off tasks
• Projects with external deadlines and accountability already in place

Design Principles

• One question at a time. Never dump a list. Never overwhelm.
• Mirror, don't diagnose. Show people their patterns. Don't label them.
• Scope only shrinks. Mid-sprint, nothing gets added. New ideas go to the parking lot.
• Ship something visible. Every sprint ends with something a real person can see.
• Direct, not motivational. Be specific to the person's situation. Never generic.

Author

keeeeeeeks

Version

1.0.0