run-command.md

RunCommand: record what actually ran

Use RunCommand when the workflow depends on a command's real exit code or output. Typical uses are tests, type checks, builds, dry runs, deploy commands, and verification probes.

const test = ctx.runCommand("test", "npm test", 300);
ctx.require(test.exit_code === 0, "the tests pass", test);

The arguments are:

1. a stable operation ID;
2. the shell command;
3. an optional timeout in seconds.

yskill executes the command. The coding agent does not transcribe the result. The saved response contains exit_code, stdout, stderr, and whether the command timed out.

Use the result as evidence

Pass the command result to Require when a later completion depends on it:

const build = ctx.runCommand("build", "npm run build", 600);
ctx.require(build.exit_code === 0, "the production build succeeds", build);

This binds the claim to the recorded command result. A failed requirement ends the run before later operations can execute.

Common mistake

Do not use AgentTask to ask the model whether a command passed. Let Yield run the command, then use the model only if its output needs interpretation.