Codebase Customization Initializer
Explore a codebase and create or update customization files that help AI coding agents work effectively.
The skill explicitly limits work to creating or modifying chat customization files and forbids starting the task named in the argument, providing basic scope control. It does not require confirmation before edits, disclose sensitive-data or external-data flows, or define backup and rollback procedures, so points are deducted.
The workflow is broadly coherent across discovery, exploration, generation/merge, and iteration. However, it depends on the unspecified agent-customization skill, optionally on chronicle, and on undefined subagent behavior, with no diagnostic handling for invalid input, permission failures, or failed exploration; the static ceiling therefore requires a conservative deduction.
The audience and core scenario are reasonably clear: produce or update repository-level customization files for AI coding agents, optionally focused on an argument. Trigger semantics, non-fit boundaries, complex-repository layering rules, and Chinese/mainland-China environment support are not sufficiently specified, so points are deducted.
The file has a name, description, argument hint, disabled model invocation, and a staged workflow, while repository context supplies MIT licensing and official provenance. It lacks versioning, changelog, explicit maintenance ownership, installation/dependency notes, examples, and troubleshooting; its search list also inconsistently names AGENT.md versus AGENTS.md, so points are deducted.
The skill could produce directly useful agent guidance by investigating build commands, architecture, conventions, and pitfalls, and it specifies a change-summary table. There are no representative outputs or result verification, and outcomes depend on the related skill and exploration quality, so static evidence supports only a limited score.
The skill text is auditable, and the repository provides general CI, test scripts, and governance material. There is no test or execution evidence for this skill itself, and the supplied CI does not establish its key-path behavior, so verifiability remains limited.
- Confirm the exact files to be created or modified before execution and retain a recoverable backup; no rollback procedure is specified.
- The referenced agent-customization and chronicle skills are not included in the evidence, so missing dependencies may prevent completion or reduce output quality.
- Resolve the AGENT.md versus AGENTS.md naming inconsistency and add failure handling, examples, versioning, and maintenance information.
- Chinese interaction and mainland-China network availability are not documented; disclose any dependence on unreachable overseas services.
What it does & when to use it
init creates or updates codebase customization files for AI coding agents, including AGENTS.md, .github/copilot-instructions.md, skills, and custom agents. It explores project structure, existing conventions, build and test information, and documentation before generating or merging guidance. The command only handles chat customization files and does not start the development task described by a user argument. It is part of the Agent Skill collection in microsoft/vscode, a repository released under the MIT license.
The command searches for existing agent instruction files, README files, and rule files, then explores the codebase for build and test commands, architecture, conventions, documentation, and development pitfalls. It creates or updates customization files for AI coding agents, preferring AGENTS.md for new files and preserving valuable content in existing AGENTS.md or .github/copilot-instructions.md files. When finished, it prints a table of added or modified files with their purpose and proposes related agent customizations to create next.
- A team maintaining a large codebase wants new AI coding agents to learn project conventions, architecture boundaries, and common development commands quickly.
- A maintainer has an outdated AGENTS.md or .github/copilot-instructions.md and wants to refresh it without discarding useful existing guidance.
- A developer starting a new project wants initial agent instructions, skills, or custom agents based on a project description.
- A team wants focused guidance for a particular area, such as testing, frontend work, or backend work, without beginning the implementation task itself.
Pros & cons
- Covers the workflow from discovering project conventions through generating or merging customization files.
- Follows a link-don't-embed approach so existing documentation is referenced rather than duplicated.
- Preserves valuable existing guidance and prefers AGENTS.md for new instruction files.
- Supports both new-project initialization and maintenance of existing agent guidance.
- The source provides no standalone installation command or host-client installation instructions.
- It relies on the related `agent-customization` skill for detailed guidance about customization file types.
- The SKILL.md sets `disable-model-invocation` to true, so the skill must be explicitly invoked rather than automatically selected.
- The source does not show a skill-specific test suite, platform coverage, or failure-recovery procedure.
How to install
The supplied README does not document a standalone installation command for this skill collection. The skill file is located at extensions/copilot/assets/prompts/skills/init/SKILL.md; use the installation process of the host Agent Skills client if applicable. No more specific installation procedure is provided by the source.
How to use
Invoke the init command explicitly. Run it without an argument for general codebase customization, or provide an optional focus area or pattern, such as init with a request to document testing conventions. The command briefly announces that it is exploring the codebase and working on customization files, then reports the files it added or modified and why they are useful.