carrel

Portable AI-assisted research vault operations for setup, document ingestion, transcription, web capture, environment repair, trust and sensitivity routing, reflection, collaborator handoff, and feedback export. This is the standalone engine for hosts without the Carrel Claude Code plugin, such as Cowork, Codex, Kimi Code, Gemini CLI, OpenCode, and Claude.ai skill uploads. Do not use when the Carrel Claude Code plugin or the installed carrel CLI is available; that plugin drives the typed CLI as its runtime, not this pack.

This skill should be used when a researcher wants to configure, understand, or adjust unattended vault maintenance. Triggers on 'schedule', 'automate', 'overnight', 'background processing', 'morning brief', 'unattended', 'what did Carrel do', 'review automation settings', 'trust level', or '/carrel-automate'.

This skill should be used when a researcher has audio or video files to transcribe, or a YouTube URL to get a transcript from. Triggers on 'transcribe', 'convert this recording', 'audio to text', 'get the transcript', or when an audio/video file or YouTube URL is provided.

This skill should be used when a researcher wants to share their Carrel vault with someone else — an RA, co-author, lab member, or advisor. Triggers on 'share with', 'onboard a collaborator', 'generate a handbook', 'lab handbook', 'help my RA get up to speed', or when the /carrel-share command is invoked. Distinct from /carrel-setup, which sets up a NEW vault for a NEW researcher; this skill assumes an existing vault with accumulated context.

This skill should be used when a researcher wants to set up their AI research environment, mentions 'setup', 'get started', 'configure', 'onboard', or when opening a project folder with no .carrel/ directory. Also triggers on 'what tools do I have', 'check my setup', 'add a new tool'.

This skill should be used when a researcher wants to build, query, or maintain a synthesized knowledge base across their sources. Triggers on 'field map', 'knowledge map', 'what do my sources say about', 'synthesize across papers', 'literature review', 'systematic review', 'wiki', 'track contradictions', 'knowledge base', 'I keep losing track', 'lint wiki', 'wiki health', or when the agent observes repeated cross-source questions in a vault with 15+ papers.

This skill should be used when a researcher wants to think through ideas, discuss a paper, get feedback on arguments, explore connections, brainstorm, or needs intellectual engagement. Also covers research self-portrait synthesis — surfacing patterns across a researcher's own recent work ('show me my research patterns', 'what have I been working on', 'give me a mirror', '/carrel-mirror'). Triggers on 'help me think', 'what do you think', 'push back', 'what am I missing', 'I'm stuck', 'explore connections', 'mirror', 'self-portrait'.

This skill should be used when evaluating an external skill, MCP, or tool for absorption into Carrel ('should we absorb this', 'evaluate for absorption', 'does this overlap'), when Claude has just created a custom artifact for a researcher and needs to log it and update their environment dashboard, when promoting a recurring custom solution to a template ('promote this to a template', 'make this official'), when walking a researcher through a plugin version upgrade ('check for updates', 'what changed', '/carrel-migrate'), or during quarterly upstream review ('review upstream sources', 'what capabilities have we added', 'check registry').

This skill should be used when a researcher wraps up a work session and wants a quick check-in on what worked and what didn't ('let's reflect', 'wrap up', 'what did we do today', '/carrel-reflect'), when the SessionEnd hook surfaces a reflection prompt, when issues from a session need to be captured to the friction log for later improvement, or when the researcher wants to share feedback with the plugin maintainer ('share my experience', 'generate feedback', '/carrel-feedback'). Distinct from research-partner (which engages with research content) — this skill is meta: reflection on the tool and workflow, not on the research itself.

This skill should be used when a researcher wants to create, search, organize, or manage notes in their Obsidian vault. Triggers on 'create a note', 'find my notes about', 'organize', 'search vault', 'link notes', 'use template', 'vault status', 'check inbox', 'what's in my vault', or any vault file operation.

This skill should be used when a researcher wants to convert a PDF, Word document, PowerPoint, spreadsheet, or image to markdown. Triggers on 'convert', 'import', 'add this paper', a dropped file path, or any mention of PDF/DOCX/PPTX/XLSX conversion.

This skill should be used when a researcher wants to add, remove, or review multi-model agent teammates (Codex/ChatGPT, Gemini, Kimi) inside Claude Code. Triggers on 'teammates', 'multi-agent', 'multi-model', 'second opinion', 'add Gemini', 'use ChatGPT here', 'orchestrate models', or when `/carrel-setup` enters Phase 5b.

Guided recovery for environment.json drift. Use when the researcher runs /carrel-fix, mentions environment drift, broken setup state, or wants help repairing conflicting settings.

This skill should be used when a researcher wants to save web content to their vault. Triggers on URLs, 'save this article', 'clip this page', 'capture this website', or 'add this to my vault' with a URL.