Saltus MCP Skill
A ready-to-use agent skill for connecting AI clients to the saltus/* MCP/Abilities surface. Drop it into your agent's skills directory and it will guide the agent through health checks, discovery, safe reads, and confirmed writes — without you repeating the integration rules.
Download
Or fetch it directly:
curl -L -o saltus-mcp-SKILL.md \
https://raw.githubusercontent.com/SaltusDev/saltus-framework/main/.agents/skills/saltus-mcp/SKILL.mdWhy use it
Saltus exposes 18 saltus/* abilities over WordPress-native MCP/Abilities. Clients that treat them as raw REST calls tend to:
- guess post type slugs instead of discovering them with
list_models - write meta before checking the normalized field paths from
list_meta_fields - retry permission failures that will not resolve until roles change
- fire broad list queries as a discovery shortcut, tripping rate limits
- force-delete without explicit confirmation
The skill encodes the guardrails the framework already expects — the health-first call flow, normalized metadata reasoning, read-confirm-write confirmation, and permission/rate-limit handling — so any agent that loads it behaves correctly out of the box.
How to install
Copy SKILL.md into your agent's skills folder. For example, for a filesystem-based agent:
mkdir -p ~/.agents/skills/saltus-mcp
curl -L -o ~/.agents/skills/saltus-mcp/SKILL.md \
https://raw.githubusercontent.com/SaltusDev/saltus-framework/main/.agents/skills/saltus-mcp/SKILL.mdThe skill front matter declares when it applies:
name: saltus-mcp
description: >-
Use when connecting an AI agent/client to Saltus Framework MCP/Abilities
(saltus/*) on an active WordPress site — discovery, health-first flow,
model/meta inspection, safe read/write patterns, permission and rate-limit
handling, and write confirmation flows.Agents that auto-load skills by description will pick it up whenever a task mentions connecting to Saltus MCP or working with saltus/* tools.
How it works
The skill is a single self-contained SKILL.md organized around the client integration guide:
- Discovery — filter abilities for the
saltus/prefix and trust the ability metadata (meta.mcp_tool,meta.namespace,meta.transport). - Health first — call
saltus/get-healthat session start and branch onstatus,error_rate, and latency. - Model and metadata discovery —
saltus/list-modelsandsaltus/list-meta-fieldsbefore touching content; prefernormalized.fieldspaths. - Safe reads — list, confirm, then single-get with small page sizes.
- Safe writes — read model/meta, build a minimal patch, summarize to the user, execute, then re-read to confirm.
- Destructive actions — require explicit confirmation for force deletes, bulk deletes, reordering, settings updates, and serialized-meta writes.
- Failure handling — never retry permission failures; respect
retry_afteron rate limits; confirm mutations against the post-write cache clear. - Planning heuristics — an intent-to-first-tools table and copy-paste prompt guidance for agents.
- Anti-patterns — the behaviors to avoid.
Example
A task like "update the director of the movie with the slug interstellar" resolves through the skill's flow:
- Call
saltus/get-health→status: ok. - Call
saltus/list-models→ find themoviepost type. - Call
saltus/get-meta-fieldswithpost_type: "movie"→ normalized pathdirectormaps to the writable meta root. - Call
saltus/list-postswithpost_type: "movie",search: "interstellar"→ confirm the post ID with the user. - Call
saltus/get-post→ read currentdirectorvalue. - Summarize: "Update
directoron post 42 (Interstellar) from 'Nolan' to 'Christopher Nolan'?" - Call
saltus/update-postwithpost_id: 42,meta: { director: "Christopher Nolan" }. - Call
saltus/get-postagain → report the confirmed value.
Keeping it in sync
- The authoritative copy lives in the repository at
.agents/skills/saltus-mcp/SKILL.md. - The docs site serves a copy at
/downloads/saltus-mcp/SKILL.mdfor direct download. - The skill mirrors Client Integration; when the client guidance changes, update both.
