[feature] Agent Context File Generation

[Kukkerem]

Goal

Add a generate-context subcommand that produces an AGENTS.md file from the CLI's
registered subcommands, parameter definitions, and optional author-defined annotations. The
file encodes usage invariants that agents cannot intuit from --help.

Motivation

Agents need invariants made explicit: which operations are destructive,
which parameters require special formatting, which resource kinds exist, and how credentials should be provided.

A context file answers these questions at near-zero token cost because it is loaded once
per conversation rather than reconstructed each time. A skill file is cheaper than a
hallucination.

Task

A generate-context subcommand needs to be added that introspects all registered subcommands
and generates structured Markdown. The output should include tool names, descriptions,
parameter tables with types and allowed values, and an authentication section listing
environment variables for sensitive params.

CLI authors should be able to attach agent-specific guidance to subcommands and parameters
(e.g. "always specify --kind to limit scope") that gets included in the generated file.
This guidance is not something the framework can infer automatically, so it requires an
annotation API.

The command should write to stdout by default, with an optional -o flag for file output.

Optionally, the command could support YAML frontmatter output for compatibility with agent
frameworks that consume structured skill files.

Acceptance criteria

• generate-context subcommand generates AGENTS.md content
• Generated content includes tool names, descriptions, parameter tables, and enum values
• CLI authors can attach agent-specific guidance to subcommands and parameters
• Sensitive params are listed in an authentication section with env var names
• Output goes to stdout by default with optional -o for file output
• Tests assert the generated markdown structure and content

[gergely-szabo-sap]

Does it make any sense to support the generation of SKILL.md files?

[Kukkerem]

It's better for token usage, and it would be great if we had some basic, multi-step use cases. I don't have too many in mind right now, but this is why it's the last issue, it could come from the users during testing. We could skip this til we have some insights.