What files does the onboarding skill prioritize?

The skill reads essential meta-documentation in a specific order, including agents.md, repo_workflow.md, onboarding.md, and REPO_MAP.md, before looking at source code.

How does this prevent context window overflow?

It uses 'bounded discovery,' which limits the AI to identifying and opening a maximum of 10 additional files per task slice instead of performing broad repository scans.

Where does the skill log its progress?

It automatically appends a structured entry to agent_logs/current.md, documenting the agent's current intent and planned next steps.

Does this skill work with any repository?

While optimized for repositories with structured documentation like TML-bench, its deterministic pattern is ideal for any project requiring strict agent state management.

Repository Onboarding Agent

Name: Repository Onboarding Agent
Author: MykolaPinchuk

byMykolaPinchuk

生产力与工作流

Streamlines AI agent onboarding to new codebases through deterministic file discovery and structured state reporting.

关于

The Repository Onboarding skill provides a deterministic, step-by-step process for Claude to understand a new repository's architecture and workflow. It prevents context bloat by reading essential index files first—such as agents.md, repo_workflow.md, and REPO_MAP.md—followed by a strictly bounded discovery of relevant source files. By producing a standardized output contract that includes current state summaries, next-step options, and intent logging, it ensures that AI agents can begin contributing to complex projects with high accuracy and minimal manual intervention.

主要功能

Deterministic index file sequencing for rapid context gathering
0 GitHub stars
Standardized state reporting with next-slice action choices
Bounded file discovery limited to 10 context-specific files
Automated intent logging to agent_logs/current.md
Constraint-based repo scanning to optimize token usage

使用场景

Quickly orienting an AI agent to the architecture of the TML-bench repository
Defining clear boundaries for file discovery to prevent context window overflow
Maintaining a structured log of agent intentions and progress across multiple sessions

关于

主要功能

Deterministic index file sequencing for rapid context gathering
0 GitHub stars
Standardized state reporting with next-slice action choices
Bounded file discovery limited to 10 context-specific files
Automated intent logging to agent_logs/current.md
Constraint-based repo scanning to optimize token usage

使用场景

Quickly orienting an AI agent to the architecture of the TML-bench repository
Defining clear boundaries for file discovery to prevent context window overflow
Maintaining a structured log of agent intentions and progress across multiple sessions