Find the open-source solution that fits.

Scan common storage consumers such as AI coding tools, package-manager caches, Docker, Xcode data, browser test runtimes, and macOS app containers. Findings are labeled SAFE, ASK, or KEEP before you choose a conservative, aggressive, or custom cleanup plan, followed by a before-and-after disk-space report.

Trigger a brief reflective mode when an answer seems technically correct but misses the larger point. Phrases such as “zoom out” or “what am I missing?” prompt Claude to examine the whole situation, which can help with strategic decisions, stuck thinking, and pre-commit plan reviews. It works as a Claude Code skill or as copied instructions for another LLM.

Capture a thought without leaving the terminal: the plugin detects the track currently playing in Spotify and appends a timestamped entry to a local Markdown file for that song. You can recall notes for the current track or search the whole archive. It uses no Spotify API, OAuth, or cloud sync; macOS and Linux are supported, while Windows is not.

Provide a type specimen and the plugin selects a similar OFL-licensed Google Font, renders and bitmap-traces 82 target glyphs, rebuilds the font tables, renames the family, and emits an OTF plus an HTML specimen page. It requires potrace, Pillow, and fonttools, focuses on Latin glyphs, and performs no hinting pass.

Name a fictional character and the skill researches available information, builds a reusable persona, and conducts dialogue in that character's voice with visible source-quality indicators. Personas can be cached for 90 days, responses can follow the user's language, and pause, resume, status, and exit commands keep the session controllable. Built-in safeguards address harmful character requests.

This collection packages three skills: a triaged six-step work process with an approval gate, a structured evaluation for external repositories, and a benchmark process for choosing production AI models. Its accompanying CLAUDE.md adds rules for evidence labels, security scope, parallel-agent isolation, and commit discipline, aimed at reducing scope drift and unapproved execution.

Clicks lets an agent record a signed task receipt after work is completed and look up another agent's reputation before delegating to it. The records are tied to ERC-8004 identities on Base. Using it requires a minted agentId, an agent wallet with a small amount of ETH for gas, and a private-key environment variable.

Provide a comma-separated company list and the skill searches official sites, brand pages, image search, and Wikipedia in parallel. It visually checks identity, current branding, colors, and whether a full wordmark is available before saving verified files locally. It does not require a logo-service API key.

The skill categorizes dated sessions, release notes, audits, and demo runs, then merges them into typed archive files and a smaller memory index. It leaves feedback, reusable references, user profiles, and active knowledge alone by default. Before writing or deleting anything, it presents a plan for confirmation and removes only true duplicates.

Running `/bro` stores repository-specific context that chat compaction can erase, including rules, architecture decisions, vocabulary, roadmap changes, and active debugging hypotheses. It separates persistent principles from thread state and daily logs, tags parallel chats with the title plus a UUID suffix, and can migrate older log layouts into the visible project `bro/` directory.

From any project folder, you can draft a single post, thread, reply, or image attachment in natural language. Claude shows the exact text and character count, asks for confirmation, then posts through the X API and returns the URL. Credentials live once in a global user file rather than each repository, and dry-run mode previews the payload without sending it.

Claude Sync tracks installed skills and commands on each machine, including versions and content hashes, then uses pull, push, status, and consolidation workflows to surface drift. It also provides a Git-backed message file for machine-to-machine coordination. MCP configuration, plugin installation, and automatic conflict resolution remain outside its current scope.

PaperJSX turns validated structured JSON into presentations, documents, spreadsheets, and PDF reports. It runs locally through the free-tier `@paperjsx/mcp-server` and `@paperjsx/*` packages, without an API key or hosted backend. The skill can be used with Claude Code, Claude.ai, Codex CLI, and Claude Desktop.

Vision Roadmap reads the codebase and documentation, asks structured questions about horizon, priorities, and constraints, then groups the product direction into a narrative roadmap. It can also create matching Linear issues through the GraphQL API. Teams without a Linear key can skip ticket creation and keep the roadmap document only.

Aside occasionally identifies a cognitive bias or psychological pattern reflected in the user's own words and cites an academic source. It is designed as a quiet mid-conversation observation rather than a diagnosis, profile, or recommendation. The skill runs locally without telemetry or external data transfer, and a rejection or stop phrase disables it for the rest of the session.

Bulk PR Review maps dependency layers and cross-PR conflicts before dispatching parallel reviewers within each layer. It processes layers sequentially, gathers verdicts such as approve, request changes, blocked by CI, or close, and posts them directly through `gh pr review`. The workflow requires an authenticated GitHub CLI and `jq`.

The skill starts from the git diff, detects the project's test framework and configuration, and reads existing tests to learn naming, fixtures, assertions, and file placement. It then adds coverage for happy paths, errors, boundaries, and authorization failures before running the suite. It does not assume a stack and needs no API credentials.

Provide an image path and a natural-language target, and the skill iteratively crops and rechecks the region before returning a bounding box, center point, confidence, and saved evidence. It can report that no target was found instead of inventing a coordinate; screenshot capture and UI interaction remain the caller’s responsibility.