What files does the Onboard skill prioritize?

It reads specific documentation in order: agents.md, repo_workflow.md, onboarding.md, HANDOFF.md, REPO_MAP.md, and the README to build a mental model of the project.

How does this prevent context window bloat?

By limiting additional file discovery to 10 relevant files per 'slice' and avoiding broad repo scans that consume tokens unnecessarily.

When should I use the 'Onboard' command?

You should use it whenever you start a new session in a repository or move to a different architectural part of the project.

Does it keep a history of its actions?

Yes, it automatically appends intents, planned actions, and unknowns to agent_logs/current.md for better session tracking.

What is the 'output contract' provided by the skill?

Every onboarding ends with a 5-bullet state summary, 3 choices for the next development slice, a list of read files, and a risk assessment.

Deterministic Repo Onboarding

Name: Deterministic Repo Onboarding
Author: MykolaPinchuk

byMykolaPinchuk

生産性とワークフロー

Streamlines codebase exploration for Claude through a deterministic sequence of document indexing and bounded discovery.

概要

The Onboard skill provides a structured framework for Claude to ingest new repositories by prioritizing essential architectural documents like REPO_MAP and onboarding guides before moving to limited, high-value file discovery. This approach prevents context bloat and excessive scanning, ensuring the AI agent gains a clear understanding of the project's current state, risks, and next steps through a standardized output contract and automated logging in the Codex workflow.

主な機能

Deterministic indexing of architectural documents (REPO_MAP, HANDOFF, etc.)
Bounded discovery limited to 10 context-specific files per task slice
Standardized output contract including state summaries and risk assessment
Automatic workflow logging to agent_logs for session continuity
Agent-centric README parsing for immediate quickstart execution
0 GitHub stars

ユースケース

Quickly orienting Claude to a new or complex repository
Handing off project context between different AI agents or human developers
Establishing a consistent project baseline before starting a new development cycle

概要

主な機能

Deterministic indexing of architectural documents (REPO_MAP, HANDOFF, etc.)
Bounded discovery limited to 10 context-specific files per task slice
Standardized output contract including state summaries and risk assessment
Automatic workflow logging to agent_logs for session continuity
Agent-centric README parsing for immediate quickstart execution
0 GitHub stars

ユースケース

Quickly orienting Claude to a new or complex repository
Handing off project context between different AI agents or human developers
Establishing a consistent project baseline before starting a new development cycle