onboard

MykolaPinchuk's avatarfrom MykolaPinchuk

Deterministic onboarding for this repo (read minimal index files first, then bounded discovery).

⭐0stars🔀0forks📁View on GitHub🕐Updated Dec 30, 2025
[Memory and Context Management]

When & Why to Use This Skill

This Claude skill facilitates deterministic repository onboarding for AI agents, ensuring efficient context acquisition through structured file reading and bounded discovery. By prioritizing essential index files and limiting exploratory scans, it optimizes token usage and provides a clear roadmap for task execution within complex codebases, making it an essential tool for maintaining agent focus and accuracy.

Use Cases

  • Rapid Project Orientation: Quickly bringing an AI agent up to speed on a new or complex repository by systematically reading core documentation like REPO_MAP.md, README.md, and onboarding guides.
  • Efficient Context Management: Preventing context window overflow and reducing token costs by restricting file discovery to a maximum of 10 highly relevant files per task slice.
  • Structured Handoff and Logging: Maintaining a clear audit trail of agent intent, actions, and identified risks in centralized logs to ensure seamless transitions between human developers and AI agents.
nameonboard
descriptionDeterministic onboarding for this repo (read minimal index files first, then bounded discovery).

When invoked (or when the user says Onboard), do this exactly:

  1. Read (in order, if present):

    • AGENTS.md (if not already in context)
    • onboarding.md
    • HANDOFF.md
    • REPO_MAP.md
    • README.md (focus on "For agents" + quickstart)

  2. Bounded discovery:

    • Identify up to 10 additional files you need for the current slice.
    • Print the list (file path + 1-line reason each).
    • Open only those files after listing them.
    • Do not perform broad repo scans.

  3. Output contract:

    • 5 bullets: current state summary
    • 3 options: next slice choices (each 1–2 bullets)
    • Files read (including the bounded discovery list)
    • Unknowns/risks (<= 5 bullets)

  4. Log:

    • Append an entry to agent_logs/current.md with:
      • intent, actions planned, and key unknowns