This lesson's interactive features are locked, please to keep using them

Dynamic Prompts

OpenCode's command pipeline supports a few useful prompt features:

Positional arguments like $1, $2, and $ARGUMENTS
Shell output injection like !`bun test`

These are command features. Invoking a skill with @skill-id activates the skill body without parsing command arguments or shell blocks (but it's still valid to invoke a skill and provide additional instructions in the same prompt).

For example, a file at .opencode/commands/linusfunc.md could contain:

---
description: Builds a function like Linus Torvalds would
---

Write a new function, complete with tests the way Linus Torvalds would. The function is $ARGUMENTS

The $ARGUMENTS variable is replaced by whatever you type after the command:

/linusfunc a sum function that adds two numbers together

You can also break the arguments into separate positional variables if you like:

The function is $1
It should $2 include tests

/linusfunc sum not

Which produces this prompt:

The function is sum
It should not include tests

File References

While using @path-to-file works in the OpenCode prompt box to select a file and include its contents, it doesn't work in a command prompt: it just stays plain text. Instead, just tell the agent to read the file:

Read `src/asteroid.py`, then make me a new asteroid with jaggier edges.

Assignment

Tell OpenCode to create .opencode/commands/playground.md. Use $ARGUMENTS for the port, defaulting to 8080, and inject the output of lsof -nP -iTCP -sTCP:LISTEN into the prompt. Have the agent check for an existing Sloperator server before starting uv run playground/server.py <port> in the background. Keep the port argument outside the shell-injection block.
Invoke /playground 8080, then invoke it again. Inspect the injected listener list and confirm the second invocation reuses the running server.

With the playground still running, tell OpenCode to run and submit the CLI tests from the project root.