Prime Session Skill Skill

Prime Session Skill

Purpose

Provide project orientation by reading key documentation files. This skill helps Claude understand the project's architecture, conventions, and current priorities before starting work. It's designed to auto-invoke when entering a new repository.

Natural Language Triggers

Automatic Invocation:

When Claude detects it's in a new/unfamiliar repository
When project context seems missing or stale
At the start of a new conversation in a repo

Explicit Invocation:

"Orient me to this project"
"What is this project?"
"Prime session"
"Get project context"
"Read the project docs"
"Familiarize yourself with this codebase"

Auto-Invoke Conditions

This skill should auto-invoke when:

New Repository Detected:
- Claude hasn't seen this repo before in session
- Git remote URL is unfamiliar
Context Seems Missing:
- Claude is unsure about project conventions
- Module emojis unknown
- Architecture unclear
User Asks Basic Questions:
- "How does this project work?"
- "What's the structure here?"
- Questions indicating lack of project context

Workflow Execution

Phase 1: Detect Repository

Get Git Context:

git remote -v
git branch --show-current
pwd

Identify Project:
- Repository name
- Current branch
- Working directory

Phase 2: Read Core Documentation

Read these files in order of priority:

CLAUDE.md (Primary)
- Project-specific Claude Code guidance
- Module definitions and emojis
- Development priorities
- Conventions and standards
README.md (Secondary)
- Project overview
- Installation/setup
- High-level architecture
Architecture Docs (If exist)
- docs/ARCHITECTURE.md
- docs/DESIGN.md
- CONTRIBUTING.md
Package/Config Files (Context)
- package.json (Node.js)
- pyproject.toml (Python)
- Cargo.toml (Rust)
- go.mod (Go)

Phase 3: Build Mental Model

From the documentation, extract:

Project Identity:
- What does this project do?
- Who is it for?
- Current development phase
Architecture:
- Key components/modules
- Directory structure
- Tech stack
Conventions:
- Module emojis (from CLAUDE.md)
- Commit message format
- Branch naming
- Code style
Priorities:
- Current development focus
- What NOT to change
- Quality standards

Phase 4: Present Orientation

Display a concise summary (don't take any action):

📍 Project: {project-name}
   Repository: {owner/repo}
   Branch: {current-branch}

📋 Overview:
   {Brief project description}

🏗️ Architecture:
   {Key components/modules}

🎯 Current Focus:
   {Development priorities from CLAUDE.md}

📦 Modules:
   - {module1} {emoji}: {description}
   - {module2} {emoji}: {description}
   ...

📝 Conventions:
   - Commits: {format}
   - Branches: {format}

✅ Ready to assist with this project.

Orientation Depth

After completing quick orientation, offer deeper exploration:

AskUserQuestion:
  question: "Quick orientation complete. Would you like more detail?"
  header: "Depth"
  options:
    - label: "This is enough"
      description: "Continue with current context"
    - label: "Deep dive"
      description: "Read all architecture docs, scan structure, review git history"
    - label: "Specific area"
      description: "Focus on a particular module or component"

Quick Orientation (Default)

Read CLAUDE.md and README.md
Extract key conventions
~30 seconds

Deep Orientation (On Request)

Read all architecture docs
Use Serena for semantic structure understanding:
- get_symbols_overview() on key source files
- find_symbol() to locate major components/modules
- Build symbol-level mental model without reading entire files
Scan directory structure (for non-code layout)
Review recent git history
Understand current work in progress

Serena Integration Benefits:

Understand code architecture without 100+ file reads
Get symbol-level structure of main components
Identify key classes, functions, types efficiently
Build precise mental model for future work

Trigger deep orientation with:

User selects "Deep dive" in AskUserQuestion
"Give me a thorough orientation"
"Deep dive into this project"
"I want to understand everything"

Silent Mode

When auto-invoking, the skill can run silently:

Read documentation without outputting summary
Store context for later use
Only show summary if user asks

Caching Behavior

To avoid re-reading on every interaction:

Session Memory:
- Remember orientation within conversation
- Don't re-read unless files changed
Staleness Detection:
- Check if CLAUDE.md was modified since last read
- Re-orient if project docs updated

Error Handling

No CLAUDE.md Found

ℹ️ No CLAUDE.md found in this project.

   This project doesn't have Claude Code-specific guidance.
   Reading README.md for basic orientation...

   Consider creating a CLAUDE.md for better assistance.

No README.md Found

⚠️ No README.md or CLAUDE.md found.

   Limited project context available.
   Scanning directory structure for orientation...

Empty Repository

ℹ️ This appears to be a new/empty repository.

   No documentation found to orient from.
   Ready to help you set up the project!

Integration with Other Skills

Provides context to:

All other skills - Module emojis, conventions
commit-changes - Commit format
setup-work - Branch naming, priorities

Auto-invokes before:

Any skill that needs project context
First interaction in a repository

Best Practices

✅ DO:

Read CLAUDE.md first (most relevant)
Keep orientation summary concise
Note current development priorities
Remember module emojis
Respect project conventions

❌ DON'T:

Take any actions (orientation only)
Modify any files
Make assumptions about undocumented conventions
Output lengthy summaries (be concise)
Re-read docs unnecessarily

Example Output

📍 Project: escapement
   Repository: fusupo/escapement
   Branch: main

📋 Overview:
   Generic Claude Code workflow system for structured development.
   Provides skills, commands, and agents for multi-project support.

🏗️ Architecture:
   - skills/     Automated workflow modules
   - commands/   Quick CLI helpers (being converted to skills)
   - agents/     Specialized subagents (extensibility ready)
   - docs/       Extended documentation

🎯 Current Focus:
   - Rearchitecting commands as skills
   - Integrating Claude Code affordances
   - Completing documentation

📦 Modules:
   - workflow 🔄: Core workflow definition
   - skills 🎯: Automated workflow skills
   - commands ⚡: Quick CLI helpers
   - docs 📚: Extended documentation
   - install 🔧: Installation and setup

📝 Conventions:
   - Commits: {emoji}{emoji} type(scope): description
   - Branches: {issue-number}-{description}

✅ Ready to assist with this project.

Version: 1.1.0 Last Updated: 2025-12-31 Maintained By: Escapement Changelog:

v1.1.0: Added AskUserQuestion for orientation depth selection
v1.0.0: Initial conversion from commands/prime-session.md

Agent Skills: Prime Session Skill

Install this agent skill to your local

Skill Files