Smoke Test

Trace a resource, event, or feature through the system and verify it works end-to-end.

Preferences

_Read ~/.claude/skills/smoke-test/preferences.md using the Read tool. If not found, no preferences are set._

Project context

_On startup, use Bash to detect: CLAUDE.md presence, project stack files (ls package.json Cargo.toml pyproject.toml go.mod requirements.txt), and current git branch. Skip any that fail._

Command routing

Check $ARGUMENTS:

• `help` → display help then stop
• `config` → interactive setup then stop
• `reset` → delete ~/.claude/skills/smoke-test/preferences.md, confirm, stop
• anything else → run smoke test

Help

Smoke Test — Trace and verify something works end-to-end

Usage:
  /smoke-test <natural language>       Build and run a trace plan
  /smoke-test config                   Set environment and verbosity defaults
  /smoke-test reset                    Clear preferences
  /smoke-test help                     This help

Examples:
  /smoke-test uploaded a file to workspace 30 in prod, check the pipeline
  /smoke-test verify login flow on staging
  /smoke-test did my migration run correctly in production?
  /smoke-test check webhook fires after order creation in dev

How it works:
  1. Reads project context (CLAUDE.md, repo structure, APIs)
  2. Builds a trace plan — ordered checks
  3. Confirms the plan with you
  4. Runs each check, reporting pass/fail
  5. Summarizes with diagnosis

Current preferences:
  (shown above under Preferences)

Config

Use `AskUserQuestion`:

Q1 — "Default environment?" (Production, Staging/Dev, Local, Ask each time (default)) Q2 — "Trace output verbosity?" (Concise, Detailed, Verbose on failure (default))

Save to ~/.claude/skills/smoke-test/preferences.md.

First-time detection

If no preferences file exists, show: "First time using /smoke-test? Run /smoke-test config to set default environment and verbosity, or continue — I'll auto-detect."

Then proceed normally.

Steps

1. Understand the project

Read project context silently:

• CLAUDE.md for conventions, endpoints, commands
• Repo structure (ls top-level)
• Tech stack from config files
• API base URLs from CLAUDE.md, .env, or config files
• Integration test files for endpoint patterns
• Available MCP tools in the session

2. Parse the request

From $ARGUMENTS, extract:

• What to verify: Resource, event, or feature to trace
• Environment: Where to check (use preference default or detect from input)
• Identifiers: IDs, names, values to trace
• Expected behavior: What "working" means

If critical info is missing, ask via AskUserQuestion.

3. Build the trace plan

Build an ordered list of checks. Each check:

• Check name: Short description
• How: Command or API call
• Expected: What passing looks like
• Depends on: Previous checks that must pass first

See reference/check-patterns.md for heuristics on what to check.

4. Confirm the plan

Smoke test plan for: [what we're verifying]
Environment: [env]

Checks:
  1. [Check name] — [how, briefly]
  2. [Check name] — [how, briefly]
  3. [Check name] — [how, briefly]

Estimated: [N] API calls

Use `AskUserQuestion`: "Run this trace plan?" (Run all, Select specific, Add a check, Skip)

5. Execute

Run each check sequentially (respecting dependencies):

[PASS] Check name
       Key details: value

[FAIL] Check name
       Expected: X
       Got: Y

[SKIP] Check name
       Depends on failed check

On failure: show expected vs actual, suggest likely cause, ask to continue or stop.

Output patterns by verbosity preference:

• Concise: pass/fail + one-liner
• Detailed: full API responses
• Verbose on failure: concise for pass, detailed for fail

6. Summary

Smoke Test Results: [what was verified]
Environment: [env]

  [PASS] X/N checks passed
  [FAIL] Y/N checks failed
  [SKIP] Z/N checks skipped

Failed:
  - [Check]: [one-line reason]

Diagnosis:
  [Root cause or pattern if identifiable]

Suggested next steps:
  - [Action per failure]

Principles

• Read-only by default: Never write, mutate, or delete. This is a tracing tool.
• Be resourceful: Use curl, gh, MCP tools, database CLIs, log files.
• Be adaptive: Every project is different. Use CLAUDE.md and repo structure.
• Ask when stuck: Rather than guess, ask. But try hard first.
• Show your work: Show commands run and explain what you're checking.