For LLMs

Short prompts, machine-readable documentation, and reusable skills for coding agents working with localhost2137.

Use this page when handing localhost2137 work to a coding agent. The prompts are deliberately short; the agent should inspect the installed runtime and plugin instead of relying on a longer prompt full of guessed details.

Copy-paste prompts

Integrate and test

Use localhost2137 to test this integration locally. Inspect the installed plugin, preserve the
application's normal service boundary, use an isolated instance, verify the behavior, and clean up.
Do not invent unsupported plugin behavior.

Diagnose a failure

Diagnose this localhost2137 integration. Separate application bugs, runtime failures, and missing
plugin behavior. Report evidence before changing code.

Build or extend a plugin

Build the smallest coherent localhost2137 plugin behavior for the application path I describe. If
that path is not explicit, ask before choosing scope. Share logic below HTTP and operation adapters,
and prove the behavior through public tests.

Add repository-specific constraints only when they materially change the work.

Give an agent the right docs

/llms.txt is the compact index. Start here when the agent can fetch more pages as needed.
/llms-full.txt contains the complete documentation corpus. Use it only when the agent cannot follow links or retrieve pages during the task.
Each documentation page also has a Markdown form: append .md to its path, or use /index.md for this site's root page.

These are machine-readable documentation endpoints. They are not skills and do not install or change anything on the developer's machine.

For a retrieval-based agent, a useful reading order is:

/index.md for the runtime and plugin boundary;
/getting-started.md for one complete loop, or the guide closest to the task;
only the concept and reference pages linked from that guide.

Do not preload the complete corpus when the agent can follow links. More context is not automatically better context.

Skills to install

Two agent skills live in the repository:

Skill	Use it for
`skills/use-localhost2137`	Configuring an installed plugin, wiring an app, and building or diagnosing integration tests.
`skills/build-localhost2137-plugin`	Authoring, extending, reviewing, and contract-testing emulator plugins.

Install only the skill that matches the task. Copy its entire directory into the skill directory documented by your coding-agent tool, including references/ and agents/. Pointing the agent at the repository's SKILL.md directly also works when the tool accepts file context.

There is no automatic skill installer, symlink behavior, or npm lifecycle hook. Installing the runtime or a plugin does not modify agent configuration.

After installation, a short prompt is enough:

Use $use-localhost2137 to integrate and test this service locally.

Use $build-localhost2137-plugin for the application behavior I describe. Ask before implementation
if the required application or SDK path is not explicit.

What a useful handoff includes

Ask the agent to report:

which application boundary now points to the emulator;
how it arranged, triggered, and observed the scenario;
how the work was verified and cleaned up;
which relevant plugin behavior is unsupported or unverified.

That is enough. Internal runtime structure belongs in the handoff only when it explains a decision or a remaining problem.