Every AI Skill in our directory that orchestrates multi-step work across tools — research pipelines, approval flows, notification routing, ETL pipelines, escalation decisions, cross-tool integrations. Each Skill has been tested end-to-end against real workflows (not toy examples). Provider-agnostic and MCP-server-aware — the Skills know which servers to call for which parts of the flow.
Custom Workflow Skills orchestrate work across multiple tools and multiple steps — read from one source, transform, decide, write to another source, notify a person, repeat. The best of these Skills pair the AI's reasoning with MCP servers for the actual work (databases, APIs, files, chat) so the AI does planning + judgment while the servers do the doing. This is where MCP shines: the Skill is the plan, the servers are the hands.
The value the Skill layer adds is reliability under complexity. Plain LLM output for multi-step work is famously fragile — one wrong assumption early cascades into a wrong result at the end. Skills enforce checkpointing: verify each step's output before proceeding, ask for human confirmation at decision points, log every action for audit. For workflows that span 5+ steps and multiple tools, this discipline is the difference between something you trust to run overnight and something you have to babysit.
What lives here vs. elsewhere: if the Skill spans multiple tools OR multiple decision points, it belongs here. If it does one focused job well (draft an email, review a contract, generate a component), it lives in its specific category. Rule of thumb: single-step Skills go in their domain category; multi-step orchestrations that combine domains go here.
Our editorial picks — heaviest tested, highest reader ratings.
The reference example of the category. Plans queries, executes across sources, cross-references, drafts brief with citations. Configurable depth. Model for how multi-step Skills should be built.
Route approval requests to the right owner based on rules. Escalates to secondary if no response in threshold time. Fully audit-logged.
Aggregate signals from multiple sources (Linear, Slack, email, PagerDuty) and route notifications intelligently — dedupe, prioritize, respect quiet hours.
Filter by sub-type, search by name, sort by recency or alphabetical.
Plan queries, execute across sources, cross-reference, draft brief with citations. Configurable depth.
Combine calendar, email triage, task list, industry news into one morning briefing. Delivered by 8am.
Pull from Linear/Jira, GitHub/GitLab, Slack activity — produce a weekly report on shipped, in-flight, blocked work.
When an incident fires: page owners, spin up war-room channel, draft status page update, coordinate resolution.
Route approval requests to right owner. Escalate to secondary if no response in threshold. Audit-logged.
Approve/reject/escalate expenses based on policy. Flags anomalies for manual review. Audit trail per decision.
Route drafts through editorial approval stages. Assigns reviewers based on content type. Tracks changes.
Aggregate signals from Linear/Slack/email/PagerDuty. Route intelligently — dedupe, prioritize, quiet hours.
Batch low-priority notifications, deliver as periodic digest. Preserves urgent for immediate delivery.
Alert on customer signals across support, product, billing tools. Route to right internal owner with context.
Ingest CSV files into a database with schema validation, dedup, and error quarantine. Idempotent.
Extract from a paginated API, transform, load to warehouse (Snowflake/BigQuery/Postgres). Retry-safe.
Sync data between databases (Postgres to Snowflake, MySQL to Postgres). Handles schema mismatches gracefully.
Extract structured data from PDFs, emails, or docs into a database or spreadsheet. Human review at low confidence.
Given a signal, decide whether to escalate and to whom. Uses org chart + severity rules + business hours.
Detect churn signals across product usage + support + billing. Escalate at-risk accounts to CS with talking points.
Triage security alerts by severity + context. False-positive filtering. Escalate real threats with pre-drafted response.
Take meeting notes, extract action items, create in Linear/Jira/Notion depending on owner + type.
When an email proposes a meeting, extract times/attendees, find free slots, create event with agenda draft.
When someone reports an issue in Slack, create the right ticket (Linear/Jira/Zendesk) with context. Reply with tracking link.
MCP servers our team pairs with Custom Workflows Skills for full workflows.
Post updates, seek approvals, run status announcements. Pairs with almost every workflow Skill.
Create tickets, transition states, query pipelines from workflow Skills.
Enterprise counterpart for teams on Atlassian.
Save briefs, reviews, and workflow outputs to Notion workspaces.
Send notifications, read incoming signals. Central for many multi-step workflows.
Query databases as one of many sources in a multi-step workflow.
Longer-form tutorials for the concepts behind these Skills.
Techniques for making long AI workflows reliable — checkpointing, verification, human-in-the-loop.
When to prototype in AI Skills vs. graduate to a dedicated automation platform.
How multi-step Skills call multiple MCP servers in sequence and pass data between them.
Head-to-head reviews that touch on Custom Workflows workflows.
Skills-based workflows vs traditional automation platform: where each wins.
For long chains of reasoning across tools, which provider stays coherent?
Skill-based multi-step vs provider-native autonomous modes.
Our sister sites cover errors and code recipes across the AI ecosystem. Bookmark them for when a Skill hits an edge case.