Have multiple agents argue different positions on a question or decision, then synthesize the strongest answer from their arguments. Use for architecture decisions, strategy choices, or any situation where you want multiple perspectives stress-tested.
Examine retention curves, cohort behavior, and churn drivers to identify why users leave and what keeps them. Use when retention is below benchmark, declining over time, or when planning engagement initiatives.
Audit documentation for freshness by checking for broken links, outdated version numbers, stale dates, references to removed files, and other signs of decay. Use when docs feel out of date or before a release.
Read, edit, and generate Microsoft Word documents. Use when the user mentions Word, .docx, document, contract, report, or wants to extract text from a doc, generate one from markdown, or apply tracked-change-style edits.
Scan the workspace for thin, generic, or placeholder context, then interview the user with targeted questions to fill the gaps. Patches the files in place. Use after /hatch, when context files feel templated, or when Claude keeps asking the same questions session after session. Re-runnable anytime to tighten an…
Define and execute customer escalation processes including severity classification, routing, communication cadences, and resolution tracking. Use when handling active escalations or building an escalation policy.
Analyze customer accounts to identify upsell, cross-sell, and expansion opportunities based on usage patterns, business signals, and competitive positioning. Use during account planning, QBR prep, or pipeline building.
Analyze actual writing across connected tools (Slack, Gmail, Notion) to create a personal voice reference file. Use when you want Claude to write like you.
Match newly-released Claude Code features to the user's actual workspace and propose specific adoption steps. Use after a /feature-watch report, after reading the Anthropic changelog, or when the user asks "what's new" or "what should I be using." For greenfield workspace setup, see workspace-onboarding.
Manage Google Workspace tools including Drive file management, Sheets data analysis, Docs drafting, Calendar scheduling, and Gmail operations. Use when working with Google tools or automating workspace tasks.
Build a customer health scoring model that quantifies risk and opportunity from usage data, support interactions, engagement patterns, and financial signals. Use when creating a health scoring framework, refining an existing model, or assessing a specific customer's health.
Walk a raw product idea through staged discovery — clarifying questions, spec doc, user flows, PRD. Use when someone has a feature idea but does not yet have requirements a squad can pick up. Stages are gated by user keywords ("spec doc", "design flows", "create prd"). For PRD generation alone, see prd-writer. For…
Validate a business idea before building anything. Uses structured questioning, market sizing, competitive analysis, and the "sell before you build" framework. Use when evaluating a new product idea, side project, or business opportunity.
One agent creates output, a separate agent validates against criteria, and they loop until the checker approves. Use for code that needs review, content that needs editing, or any output where quality must be verified independently.
Prepare a one-page brief before any meeting. Adapts structure based on meeting type. Use before discovery calls, demos, negotiations, renewals, or standups.
Extract key decisions, action items, and insights from meeting notes or transcripts. Use when processing meeting recordings, organizing discussion notes, or creating follow-up documents.
Audit your workspace's model usage and recommend the right model for each agent, command, and task type. Balances capability against cost. Use when setting up a new workspace, reviewing costs, or after Claude releases new models.
Route orphan files in outputs/ to the correct subdirectory and extract scattered TODOs across notes. Use to clean up an accumulated backlog of loose files and unfiled tasks. For session-level capture, see session-debrief.