docs: add SAM agent skill

[AsyncAssassin]

Summary

• Add a root-level SKILL.md for agents working with SAM (Sovereign Agent Mesh).
• Document the local sam-node MCP endpoint and the mesh inspection, service discovery, tool discovery, describe, and call workflow.
• Include safety guidance for non-loopback endpoints, sensitive data, side-effecting tools, and AWS SAM ambiguity.

Testing

• Parsed the SKILL.md frontmatter.
• Ran the documented MCP workflow against a local Docker SAM mesh with a remote calculator MCP service.
• Include safety guidance for non-loopback endpoints, sensitive data, and side-effecting tools, and explicitly exclude AWS SAM / Serverless Application Model tasks.

Closes #62.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[gemini-code-assist]

Code Review

This pull request introduces SKILL.md, which defines the SAM Agent Skill for interacting with a local sam-node MCP server, covering connection, mesh inspection, discovery, and safety guidelines. The review feedback suggests simplifying the instructions around namespaced tools returned by find_remote_tools to clarify that all returned tools are namespaced and must be described before calling.

[kaisoz]

Since this is a skill to be delivered to agents, let's move it to agents/skills. The agents folder can contain all types of documents intended to be consumed by agents (e.g. prompts)

[kaisoz]

Drop this section. The model has no capabilities to configure the harness (it'd need a tool for that). The skill should assume that the SAM node tools are already callable and only cover what to do with them

[kaisoz]

I don't think this description is accurate enough. For an agent to choose the skill, the task would already have to be framed as "a SAM task", but an agent that doesn't yet know it should reach into a mesh won't phrase it that way.

Let's make the description trigger on intent, not the name "SAM", something like

Use when a task needs a tool or capability that isn't available locally and a sam-node MCP server is connected — reach remote agents on a SAM (Sovereign A gent Mesh) network: inspect mesh state, discover reachable services, describe namespaced remote MCP tools, and call them. Not for AWS SAM / Serverless Application Model

WDYT?