Github Stars Organizer

GitHub Stars Organizer

Organize starred repositories as a repeatable workflow: export first, classify by theme, propose cleanup candidates, sync the final groups into GitHub Lists, then optionally prune old lists. Prefer usage-based and topic-based buckets over language-based buckets unless the user explicitly wants language buckets.

Auth And Safety

• Ask whether the user wants analysis only or direct GitHub changes.
• Use a classic PAT in GITHUB_TOKEN.
• Prefer user + public_repo when the session may create Lists, rename Lists, delete Lists, or unstar repos.
• Do not rely on fine-grained PATs for bulk star cleanup; they can fail on public repos not owned by the user.
• Export a backup before any destructive change.
• Require explicit approval before unstarring repos or deleting old Lists.
• Tell the user to revoke the token after the work is done.

Quick Start

1. Export the current stars with scripts/export_stars.py.
2. Classify the exported repos into 5-12 high-signal topic buckets and create:

• one JSON spec file for the target Lists
• one owner/repo text file per target List

1. If the user wants cleanup, produce High-confidence and Medium-confidence unstar candidates first.
2. Only after approval, run scripts/unstar_repos.py.
3. Sync the final groups into GitHub Lists with scripts/sync_github_lists.py.
4. If the user wants to keep only the new final scheme, run scripts/delete_old_lists.py.

Workflow

1. Export and snapshot

• Run scripts/export_stars.py before cleanup or relisting work.
• Keep both JSON and CSV backups when the user may want a spreadsheet review or rollback reference.
• Write artifacts into a user-specific working directory instead of the skill folder.

2. Classify by usage

• Group repos by how the user actually uses them: embedded, AI infra, video tools, frontend build, robotics, self-hosted infra, and similar topical buckets.
• Avoid overly granular public Lists. If a bucket is weak or mixed, keep it local-only instead of forcing a public List.
• Local-only holding buckets such as 待看, Learning, Awesome, or Papers are fine during analysis, but usually should not become final public Lists.
• Build the spec and text files described in references/spec-format.md.

3. Propose cleanup candidates

• Split unstar suggestions into at least two levels such as High-confidence and Medium-confidence.
• Show candidates before changing GitHub state.
• After approval, run scripts/unstar_repos.py against a text file of owner/repo lines.

4. Sync GitHub Lists

• Use scripts/sync_github_lists.py with the spec and list directory.
• Run a dry-run first when the account already has Lists or when the spec reuses existing list names.
• Use reuse_from in the spec when the user is already close to GitHub's 32-list cap.
• Use --enforce-exact-targets when the final Lists should exactly match the local plan, including removing leftover members from reused Lists.
• Preserve existing non-target list memberships by default.

5. Prune old Lists

• Only delete old Lists after the new final Lists are verified.
• Run scripts/delete_old_lists.py with the same spec so only the target List names are kept.
• Deleting a GitHub List does not unstar the repos inside it.

Resources

• scripts/export_stars.py: export the authenticated user's stars to JSON and CSV.
• scripts/unstar_repos.py: bulk-unstar repos from a text file after approval.
• scripts/sync_github_lists.py: create, update, reuse, and populate GitHub Lists from a local plan.
• scripts/delete_old_lists.py: delete non-final GitHub Lists while keeping the new target Lists.
• references/workflow.md: end-to-end operating procedure, guardrails, and command patterns.
• references/spec-format.md: required JSON spec format and per-list text file format.
• references/example-spec.json: reusable example of a nine-list theme-based setup.

Validation

• Prefer a dry-run before mutating GitHub.
• Compare verified List counts against the local text files after sync.
• Save machine-readable reports next to the working artifacts.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Cuopipyyy
• Source: Cuopipyyy/skillsfor_agent
• License: MIT

Install and usage instructions live in the source repository linked above.