git:20261025.a974079 to git:20261025.2e203a9

---

name: smoke

description: "Smoke-test every repo task and CLI command; warnings count as failures."

license: MIT

metadata:

kind: task

author: Médéric HURIER (Fmind)

source: github.com/fmind/dot/tree/main/skills/smoke

created: "2026-10-04"

updated: "2026-10-25"

---

# Smoke

Prove that everything a repository exposes still runs: each task, each CLI command, and each documented example, with a matrix the user can trust. Warnings count as failures. [quality-assurance](../quality-assurance/SKILL.md) owns risk-based campaigns over user journeys; [repository-review](../repository-review/SKILL.md) owns code findings. Fix defects only when the request includes fixing them.

## Workflow

1. **Baseline and isolation**: record `git status --short`. Tasks that format, lock, or generate files mutate the tree; with unrelated work present, run them in a snapshot per [git-worktree](../git-delivery/references/git-worktree/GUIDE.md) and report their diff instead of leaving it behind.

1. **Discover**: list every entry point and read each command before running it; never classify by name alone.

```bash

mise tasks ls --local --json --hidden | jq -r '.[] | [.name, ((.run // []) | join(" && "))] | @tsv'

MISE_TASK_QUIET=false mise run -n <task> # print the resolved command; quiet settings hide it

jq -r '.scripts // {} | keys[]' package.json

just --summary; make -pRrq : 2>/dev/null | grep -E '^[a-zA-Z0-9_.-]+:' | cut -d: -f1 | sort -u

```

Add CLI entry points from `[project.scripts]`, `bin` fields, and `scripts/`, plus the commands quoted in `README.md` and CI workflows.

1. **Classify**: _safe_ (check, lint, test, local build, docs build), _mutating_ (format, fix, lock, generate: snapshot only), _service_ (serve, dev, watch: start under a timeout, probe health, stop it), and _external_ (deploy, release, publish, push, upgrade, login, prune, paid or networked writes). Run an external task only with explicit authority and its `--dry-run` when it has one; otherwise list it as skipped with the reason.

1. **Run tasks**: start with the repository's aggregate gate (`mise run all`, `check`, `test`), then each remaining task individually so failures stay attributable. Close stdin, bound time, keep logs, and force tasks whose cached outputs would skip them.

```bash

mise run --force --timeout 10m <task> </dev/null >"$log" 2>&1; echo "exit=$?"

```

1. **Walk CLIs**: run `--help` on every command and subcommand recursively (Typer and Click list them under `Commands:`), then each read-only command (`status`, `list`, `show`, `doctor`, `validate`, any `--dry-run`) with representative arguments. Parse `--json` output with `jq`; a malformed document is a failure. Error paths count too: a bad argument must exit nonzero with a clear message, not a traceback.

1. **Treat warnings as failures**: scan each log for `warning`, `deprecat`, `WARN`, pytest's warnings summary, `Traceback`, and skipped tests; a warning passes only with a tracked, explained exception. Look at intentionally quiet tasks as well; quiet output must not hide a nonzero exit.

## Gotchas

- **Interactive prompts hang without a terminal**: closed stdin plus `--timeout` turns them into a recorded failure; prefer the tool's `--yes` only when its effect is safe.

- **Dependencies run twice**: `depends` re-runs prerequisites; reuse a passing result when inputs did not change instead of serial re-execution.

- **Run with the repository's default environment**: mise `[env]`, `.env`, and `CI=true` change behavior; run with the repository's defaults and note any variable you set.

- **Skipped is not passed**: a task skipped for a missing optional tool is "not run", not "passed".

## Documentation

- [mise run](https://mise.jdx.dev/cli/run.html) · [mise tasks](https://mise.jdx.dev/tasks/)

- Companion skills: [mise](../mise/SKILL.md) (task vocabulary), [cli-development](../cli-development/SKILL.md) (CLI contracts), [systematic-debugging](../systematic-debugging/SKILL.md) (root causes).