Skills Overview

Skills are the connectors between your workflows and external services. Each skill is a logical group of tools that share configuration — set the required environment variables and the skill is ready. No boilerplate, no adapters, no code.

How skills work

SWEny workflows are directed acyclic graphs (DAGs). At each node in the graph, you declare which skills are available. When the executor reaches that node, it gathers the tools from those skills and hands them to Claude. Claude calls the tools as needed to complete the node’s instruction, then the executor evaluates edge conditions to decide what runs next.

nodes:
  gather:
    name: Gather Context
    instruction: Pull error details and recent commits...
    skills: [github, sentry, datadog]

In this example, the gather node has access to all tools from the GitHub, Sentry, and Datadog skills. Claude can search code, query errors, and pull log data — all within a single node.

Auto-configuration

Skills resolve their configuration from environment variables. If the required env vars are set, the skill is available. There is no registration step.

# Set these and the GitHub + Sentry skills are ready
export GITHUB_TOKEN="ghp_..."
export SENTRY_AUTH_TOKEN="sntrys_..."
export SENTRY_ORG="my-org"

The configuredSkills() helper returns only the skills whose required env vars are present. The executor uses this to build the tool set for each node at runtime.

Skill categories

Every skill belongs to a category. Categories are used for validation — the executor checks that each node has at least one configured skill per category it needs.

Category	Purpose	Skills
git	Source control	GitHub
observability	Logs, errors, metrics, incidents	Sentry, Datadog, BetterStack
tasks	Issue tracking	Linear
notification	Alerting the team	Slack, Notification
data	Embeddings, vector stores, databases, ETL	(custom skills)
general	Catch-all when no other category fits	Supabase, (custom skills)

Built-in skills

ID	Name	Category	Required env vars
github	GitHub	git	GITHUB_TOKEN
linear	Linear	tasks	LINEAR_API_KEY
sentry	Sentry	observability	SENTRY_AUTH_TOKEN, SENTRY_ORG
datadog	Datadog	observability	DD_API_KEY, DD_APP_KEY
betterstack	BetterStack	observability	BETTERSTACK_API_TOKEN, BETTERSTACK_QUERY_ENDPOINT, BETTERSTACK_QUERY_USERNAME, BETTERSTACK_QUERY_PASSWORD
supabase	Supabase	general	SUPABASE_URL, SUPABASE_SERVICE_ROLE_KEY
slack	Slack	notification	SLACK_WEBHOOK_URL or SLACK_BOT_TOKEN
notification	Notification	notification	DISCORD_WEBHOOK_URL, TEAMS_WEBHOOK_URL, NOTIFICATION_WEBHOOK_URL, or SMTP_URL

Validation

Before executing a workflow, SWEny validates that every node’s skill requirements can be satisfied. For each node, it groups the declared skills by category and checks that at least one skill per category is configured.

Missing notification skills produce warnings. Missing skills in other categories produce errors that block execution.

Run sweny check to verify your configuration without executing a workflow.

npx sweny check --workflow triage

Skills in code

import { github, sentry, slack, createSkillMap, configuredSkills } from "@sweny-ai/core";

// Explicit skill map
const skills = createSkillMap([github, sentry, slack]);

// Or auto-detect from environment
const skills = createSkillMap(configuredSkills());

await execute(workflow, input, { skills, claude });

Custom Skills

In addition to built-in skills, you can create custom instruction skills and MCP-backed skills. See the Custom Skills guide for details on:

• Creating SKILL.md files with instruction content
• Declaring MCP server integrations
• Inline workflow skill definitions
• Multi-harness skill discovery