Publish only a reviewed procedure
Supply durable instructions and optional typed components. Keep credentials and private material out of the skill body; the tool stores the procedure in the authenticated personal or team scope.
Public MCP tool
save_cloud_skill
Purpose
Create or update a reusable Cloud Skill and produce the next immutable revision of its instructions and typed components.
When to use
- Publish a durable procedure, rule, or SOP that connected agents can recall and execute through the existing Cloud Skill contract.
- Create a new personal skill or submit a team-scope revision for the team's review flow.
- Replace an existing slug when the procedure needs a new version with a new canonical hash.
When NOT to use
- Do not store transient chat, credentials, bearer tokens, private keys, or raw secrets as skill instructions or component content.
- Do not use it for an ordinary durable fact; use remember or the SDK capture workflow instead.
- Do not assume team publish intent is immediately live: team revisions are proposed rather than directly published.
- Do not exceed the instruction, component, or component-count limits; validation rejects oversized or unsafe logical paths.
Required scope
- memory:write is required on the MCP surface.
- The tool is catalogued on the advanced surface with a pro minimum tier; the authenticated personal or team scope determines where the skill is stored.
Parameters
| Name | Type | Description |
|---|---|---|
name | string · required | Human-readable skill name; empty names are rejected. |
instructions | string · required | Root procedure body; required and limited to 64 KiB after UTF-8 encoding. |
slug | string · default: derived from name | Optional lowercase skill identifier matching the 1–64 character slug contract. |
description | string · default: empty | Optional discovery text returned by Level 0 query discovery. |
publish | boolean · default: true | Publish intent; personal scope can publish directly, while team scope creates a proposed revision. |
components_json | JSON array string · default: [] | Optional component objects with type, logical_path, content or artifact_ref, mime_type, and metadata; at most 50 and each content body is limited to 128 KiB. |
Return schema
A JSON CloudSkillResponseEnvelope with object='cloud_skill', schema_version='xmemo.cloud_skill.v1', kind='root', render_status, warnings, and data containing skill_id, slug, name, description, revision_id, revision_number, instructions, canonical_hash, and a typed resources manifest.
Examples
save_cloud_skill({
name: "deployment-procedure",
instructions: "Check the release, run the smoke test, and report the result.",
slug: "deployment-procedure",
description: "A bounded deployment verification procedure",
publish: true
})Errors
- Authorization failure: the MCP token lacks memory:write or the caller cannot write the resolved personal/team scope.
- Validation failure: name or instructions is empty, the slug is invalid, or the instruction body exceeds 64 KiB.
- Component validation failure: components_json is not a JSON array, a component path is empty/absolute/traversal-like/reserved/duplicated, a type is invalid, or a component exceeds 128 KiB.
- Scope lifecycle: personal publish is direct; team publish is stored as proposed and requires the team review flow.
Related tools
- recall_cloud_skill — Discover the skill, load its root manifest, or read a pinned resource after saving.
- execute_cloud_skill — Run a declared script component only after recalling the intended revision.
- remember — Use ordinary durable memory when a reusable procedure is not needed.
Source: Cloud Skills