# Smoke Task

> Execute an arbitrary prose task as a headless Claude session. For smoke-test pipeline use only.

- Skill: `majiayu000/smoke-task` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds add majiayu000/smoke-task`
- Raw SKILL.md: https://api.skillmd.com/api/skills/majiayu000/smoke-task/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: DevOps & Infra
- Author: majiayu000 (https://skillmd.com/u/majiayu000)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/majiayu000/smoke-task

---


# Smoke Task Skill

Execute an arbitrary prose task as a headless Claude session. For smoke-test pipeline use only.

## Usage

    /autoskillit:smoke-task <prose description of the task>

## When to Use

Used exclusively by `recipes/smoke-test.yaml` to test that the headless runner can execute
arbitrary tasks and produce capturable outputs. Not for production pipeline use.

## Critical Constraints

**NEVER:**
- Use this skill outside of smoke-test pipeline execution
- Ignore explicit output-line instructions (e.g. `key=value`) — they must be emitted exactly

**ALWAYS:**
- Execute the prose task exactly as specified in the argument
- Output any requested key=value lines on their own lines for recipe capture

## Instructions

Execute the task described in the argument provided to this skill invocation.
Complete the task exactly as specified. If the task asks you to output a specific line
(e.g. `key=value`), output it on its own line so it can be captured.