You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Remove the .agr.json metadata file that agr writes inside each installed skill directory. All metadata it currently tracks (ownership, content hash, source, tool) should be derived from agr.lock instead.
Why
.agr.json pollutes skill directories. Many users don't gitignore .claude/skills/ etc., so agr's metadata file shows up in their version control. The lockfile (#497) will track the same information at the project root, making per-directory markers redundant.
Scope
Stop writing .agr.json in fetcher.py / metadata.py
Migrate all reads of .agr.json to use lockfile data instead
Update _find_existing_skill_dir(), is_skill_installed(), conflict detection to use lockfile
Add migration: read existing .agr.json files into lockfile on first agr sync, then delete them
Update SDK Skill.content_hash to read from lockfile
Summary
Remove the
.agr.jsonmetadata file that agr writes inside each installed skill directory. All metadata it currently tracks (ownership, content hash, source, tool) should be derived fromagr.lockinstead.Why
.agr.jsonpollutes skill directories. Many users don't gitignore.claude/skills/etc., so agr's metadata file shows up in their version control. The lockfile (#497) will track the same information at the project root, making per-directory markers redundant.Scope
.agr.jsoninfetcher.py/metadata.py.agr.jsonto use lockfile data instead_find_existing_skill_dir(),is_skill_installed(), conflict detection to use lockfile.agr.jsonfiles into lockfile on firstagr sync, then delete themSkill.content_hashto read from lockfile🤖 Generated with Claude Code