auto-dot-claude
auto-dot-claude incrementally initializes a project-local .claude/ workspace.
It is not a destructive bootstrapper. It should merge, scaffold, and fill gaps. It should not overwrite project-specific files that already exist.
When to use
Use this skill when the user wants any of these:
- initialize a new
.claude/directory in a repo - normalize a messy
.claude/layout into a standard shape - add missing
skills / plans / logs / rules / settingsstructure - backfill starter files such as
CLAUDE.md,settings.json, or naming docs - keep a repo's
.claude/workspace evolving incrementally over time
Non-destructive rule
Default behavior is always:
- create missing directories
- create missing files
- append or merge when safe
- preserve existing project content
Never:
- replace existing rules just because a template exists
- rewrite an existing skill folder wholesale
- wipe user/project local settings
- silently delete unknown custom files
If a file already exists, prefer:
- keep as-is
- add a sibling example/template file
- append a clearly delimited missing section
- stop and report a conflict when merging would be ambiguous
Standard target shape
Use references/standard-dot-claude-layout.md as the default target layout.
The high-level structure has two layers:
- official-first:
CLAUDE.md- optional
CLAUDE.local.md .claude/settings.json.claude/settings.local.json.claude/commands/.claude/agents/
- project-recommended:
.claude/rules/.claude/skills/.claude/plans/.claude/logs/.claude/memory/
Not every project needs every directory on day one. Initialize the smallest useful subset.
Workflow
1. Audit first
Before creating anything, inspect:
- whether
.claude/already exists - which standard directories already exist
- which files are clearly project-specific and should be preserved
- whether the repo already has
CLAUDE.mdor similar guidance files
2. Compare against the standard shape
Compute:
- present directories
- missing directories
- present files
- missing files
- risky collisions
3. Apply incrementally
Create only what is missing.
Typical order:
- top-level guidance file(s)
.claude/settings.jsonexample or stub.claude/agents/or.claude/commands/- project-recommended extensions such as
.claude/rules/ .claude/skills/.claude/plans/.claude/logs/
4. Preserve custom work
When the repo already contains:
- custom rules
- custom skills
- custom settings
- local naming conventions
do not replace them.
Instead:
- keep them
- add missing complementary files
- optionally add a
NAMING.mdor layout note if the repo lacks one
5. Optional evolution mode
If the user wants ongoing improvement, this skill may also:
- add new recommended
.claude/subdirectories later - tighten naming conventions
- add starter templates for plans/logs
- record what was initialized and what remains manual
This should still be non-destructive.
6. Optional download mode
Download/external fetch behavior is optional, not default.
Only use it when the user explicitly wants:
- a starter template fetched from another repo
- a shared team standard synced in
- a reference file copied from an approved source
Even then:
- fetch into a temporary or clearly named destination
- review before merging
- never overwrite project files blindly
Recommended outputs
When this skill runs, it should leave behind:
- the created directories/files
- a short summary of what already existed
- a short summary of what was added
- a short summary of what was intentionally not changed
If useful, also create:
.claude/NAMING.md.claude/plans/<skill>/....claude/logs/<skill>/...
Good defaults
For this repo style, a good default is:
- keep reusable skill definitions in
.claude/skills/ - keep project planning material in
.claude/plans/ - keep raw logs in
.claude/logs/
Example invocation
请使用 auto-dot-claude,为当前项目增量初始化标准 .claude 结构。
先审计现有内容,不要覆盖已有 rules/skills/settings;
只补齐缺失目录与模板,并总结哪些内容保留不动。