Reference

recall_cloud_skill

Discover Cloud Skills, load a root manifest, or fetch one resource from an explicitly pinned revision.

Load only the layer you need

Start with query discovery, then use the returned skill or revision identifiers for root and resource recall. Level 2 resource reads require an explicit revision_id.

Public MCP tool

recall_cloud_skill

Purpose

Progressively discover Cloud Skills, recall one root revision and manifest, or fetch one resource from an explicitly pinned revision.

When to use

Start with a query when the agent needs to discover a reusable published skill by name, slug, or description terms.
Load the root instructions and resource manifest before deciding whether a component is needed.
Fetch one exact reference, example, template, document, artifact, or script from the revision selected at Level 1.

When NOT to use

Do not use a Level 2 resource request without the revision_id returned by the root recall; the service rejects unpinned resource reads.
Do not treat a descriptor from Level 0 as the skill instructions; it is only discovery metadata.
Do not use it to execute a script; use execute_cloud_skill after verifying the pinned revision and script path.
Do not expect archived skills or skills outside the authenticated personal/team scope to be returned.

Required scope

memory:read is required on the MCP surface.
The advanced tool resolves skills inside the authenticated personal or team scope and does not widen that boundary.

Parameters

Name	Type	Description
`query`	string · default: empty	Level 0 discovery text; used when skill_id and slug are empty.
`skill_id`	string · default: empty	Stable Cloud Skill identity for root or resource recall.
`slug`	string · default: empty	Human-readable skill identity for root or resource recall.
`revision_id`	string · default: empty	Optional root revision pin; required for Level 2 resource recall.
`resource`	string · default: empty	Logical component path or component ID for Level 2 recall.

Return schema

A JSON CloudSkillResponseEnvelope: Level 0 kind='descriptor' with descriptor data; Level 1 kind='root' with instructions, revision metadata, canonical_hash, warnings, and resources; Level 2 kind='resource' with the pinned component content/hash; failures use kind='error' and CloudSkillErrorPayload.

Examples

Discover a published skill

recall_cloud_skill({ query: "deployment verification" })

Load a pinned script resource

recall_cloud_skill({
  slug: "deployment-procedure",
  revision_id: "revision-id-from-root-recall",
  resource: "scripts/check.py"
})

Errors

ACCESS_DENIED: the requested skill ID does not belong to the authenticated personal/team scope.
SKILL_NOT_FOUND or SKILL_ARCHIVED: no active skill matches the supplied identity.
REVISION_NOT_FOUND: the requested or published revision cannot be resolved for the skill.
REVISION_PINNING_VIOLATION: a Level 2 resource request omitted revision_id.
RESOURCE_NOT_FOUND: the logical resource path or component ID is not present in the pinned revision.

save_cloud_skill — Create the durable skill or publish its next revision.
execute_cloud_skill — Execute a recalled script component through the sandbox.
recall_context — Use memory context retrieval when the target is agent memory rather than a Cloud Skill resource.

Source: Cloud Skills