environment-repair.md

Tutorial: resume after an environment repair

Some workflows need a person to change something outside the agent session. The run should wait without losing its earlier evidence.

The Python environment-doctor example follows this path:

1. run_command records the current probe.
2. agent_task interprets the output.
3. If unhealthy, ask_user explains the fix and waits.
4. A second run_command checks the environment again.
5. require prevents completion unless the recheck succeeds.

probe = ctx.run_command("probe-python", "python3 --version", timeout_seconds=60)
diagnosis = ctx.agent_task("diagnose", instruction, context=probe_context, schema=schema)

if not diagnosis["healthy"]:
    answer = ctx.ask_user("apply-fix", diagnosis["fix_hint"], options=options)
    if answer != "done":
        ctx.blocked("the environment fix was not applied")
    recheck = ctx.run_command("recheck-python", "python3 --version", timeout_seconds=60)
    ctx.require(recheck.exit_code == 0, "the environment probe passes", recheck)

Run the complete example:

yskill test examples/env-doctor

Source: examples/env-doctor/main.py.