A Reddit post announcing 'unstuck', a Claude Code mod that watches for loops while they happen and notifies the user in plain language.
✳︎ I made a Claude Code mod that notices when ...
└─ Make every conversation easier
352 mods for Claude Code workflow mods, ordered by the votes people give them on ClaudeMods. A good place to start: ruflo-adr,ruflo-loop-workers or code-modernization.
Each entry links to its original repository or post, and shows how to install it.
352A Reddit post announcing 'unstuck', a Claude Code mod that watches for loops while they happen and notifies the user in plain language.
Widgets docked below the prompt bar or on the side in fullscreen mode; each widget is a mod that can be toggled via slash commands, including a context window bar, usage bar, and file tree.

See the decisions Claude made in each turn, why it made them, and steer future turns with keep and avoid rules.

Track what your AI coding tools cost
syrtis.nyanako.comVisit sponsor ↗Introducing a Claude Code mod that automatically selects effort (inference level) for each request.
Provides a unified quick menu for Claude Code: displays commands and settings from various plugins, as well as Claude Code's own /config settings, in the session panel and prompt bar. It is collapsible, favorite-able, filterable, and editable in place.
One key for 140+ AI models
bazaarlink.aiVisit sponsor ↗An Accounts plugin for SimpleCORE Claude Mods that switches among Claude accounts, shows usage and storage, and adds a status band above the prompt.
Claude Code mod: tracks background services started by Claude through Bash (docker compose, colima, brew services and launchctl), and lets you stop or keep them from the /services pane.
The author released effort-cycle, a plugin that uses Alt+E / Alt+Shift+E to cycle Claude Code effort levels (low→max) and shows a colored level bar at the bottom.
# harness-scope The Claude Code Mod itself: the folder Claude Code installs. It turns your global skills, agents, instruction files (CLAUDE.md and rules) and tools on or off per repo, using named profiles you keep in `~/.claude/harness-scope/profiles/`. A repo picks one profile with a one-line `.claude/harness-scope.json`; the repo's own skills, agents and CLAUDE.md always stay.
Puts `~/.claude` under version control, so trying a plugin, mod or skill is reversible: take a snapshot, experiment, roll back if you do not like the result.