Skills

Eight skills, shared and host-neutral. Claude Code loads them from the plugin; every other host reads them from .agents/skills/. Each has a SKILL.md the agent selects from its description, so describing the task is usually enough.

Skill	Primary job	Default effect	What you receive
mule-api-design	Shape or assess an HTTP API contract	Read-only workshop or assessment; writes local RAML/OAS only when authoring is requested; Design Center mutations are separate	Decision ledger, operation model, contract or findings, and validation evidence
mule-docs	Explain the current Mule project for a named audience	Writes only the requested documentation scope	Evidence-backed Markdown/Mermaid, source map, and clearly labeled gaps
mule-development	Implement Mule XML, DataWeave, APIKit, connector, or configuration behavior	Changes production source/configuration inside the requested scope	Implemented behavior, changed-file summary, validation evidence, and remaining gaps
mule-testing	Author or repair behavior-focused MUnit coverage	Changes tests, fixtures, mocks, and test-only configuration	Test intent, focused/full results, and any unproven behavior
mule-troubleshooting	Diagnose incidents and failures	Read-only diagnosis by default	Root-cause assessment or ranked hypotheses, confidence, evidence gaps, and fix plan
mule-ops	Assess runtime health and deployment evidence	Read-only telemetry analysis by default	Health assessment, evidence coverage, risks, and recommended actions
mule-review	Review changes, repositories, PRs, or release readiness	Read-only; no implementation, comment, or PR-state change unless requested	Prioritized findings, fix options, coverage limits, and a readiness verdict when applicable
mule-build	Run local readiness, security, MUnit, packaging, and release-preparation gates	Creates target/ artifacts during packaging; source/Git changes require an explicit version or release request	Exact checks, test totals, artifact or failure evidence, and release actions performed

See Mule Skills in action for an evidence-backed change journey and a read-only risk review. The examples deliberately show a blocked test gate rather than presenting a tests-skipped package as a successful release.

Choosing the right skill

Request	Start with	Add when needed
Design or assess an HTTP API contract	mule-api-design	Development for Mule/APIKit implementation after approval
Explain or refresh the project	mule-docs	Optional business-context questions when source cannot establish purpose or ownership
Implement a change	mule-development	Documentation refresh, then change review
Add or repair MUnit tests	mule-testing	mule-build for execution; development if a product defect is exposed
Diagnose a symptom	mule-troubleshooting	mule-ops for authorized runtime evidence; development only when a fix is requested
Assess current runtime health	mule-ops	mule-troubleshooting when a specific causal question emerges
Review a change or repository	mule-review	mule-ops only for authorized, material runtime verification
Prepare a release	mule-build	Release-readiness review before commit/tag; mule-ops for separately approved publish/deploy

What they read

Every skill reads AGENTS.md for shared project context, which is why project setup matters more than the install itself. Skills refer to their bundled references and scripts through <skill-root> and <skills-root> placeholders, so the same file works under a plugin install and a vendored install.

Documentation and review questions are optional and non-blocking. When business information would materially improve the result, the skill offers concise choices plus Other and Not sure / Skip, then continues with verified technical evidence if you skip.

Runtime evidence is gated, not assumed

mule-api-design, mule-ops, mule-troubleshooting, and mule-review confirm Anypoint access before their first connector call, and offer setup, supplied exports, or a repository-only scope when it is missing. See Anypoint access.

Example prompts

Use mule-docs to refresh architecture and operations documentation. Ask optional business questions
with choices where the repository cannot establish important context.

Use mule-api-design to turn these requirements into a consumer-centered RAML or OAS contract.

Use mule-development to implement this Mule change and complete the post-development checklist.

Use mule-testing to add faithful MUnit coverage for this behavior, then run the focused and required
full tests without weakening assertions or mocks.

Use mule-troubleshooting to find the root cause of the timeouts in this flow, then stop before
changing anything.

Use mule-ops to check this application's health for the last six hours. If Anypoint access is not
configured, tell me what you need instead of guessing.

Use mule-review in change-review mode for this branch. Report findings and fix options; do not
modify source or PR state.

Use mule-build to validate and package this application without performing any release action.

Under a Claude Code plugin install, skills are namespaced, so mule-skills:mule-review invokes one directly.