Skip to content
Back to skills

El Vaporub

ASecurity

Use at the very start of a failure investigation, before forming any theory, to rule out your own stale state — unsaved file, unrestarted process, stale cache, wrong branch or worktree, wrong file, outdated build output. Run the ladder once. NOT for an unavailable or misconfigured environment; use ya-comiste.

  • 2 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 6, 2026
ai-agentsgo

Works with

  • cli

Security analysis

A100/100

Scanned September 6, 2026

npx -y skills add mcasillas17/mexican-mom --skill el-vaporub --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of El Vaporub?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for El Vaporub
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/mcasillas17-el-vaporub/badge)](https://www.skillsdirectory.com/skills/mcasillas17-el-vaporub)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: el-vaporub
description: Use at the very start of a failure investigation, before forming any theory, to rule out your own stale state — unsaved file, unrestarted process, stale cache, wrong branch or worktree, wrong file, outdated build output. Run the ladder once. NOT for an unavailable or misconfigured environment; use ya-comiste.
---

# ¿Te duele? Ponte Vaporub

Before you theorize, rule out the mess you made yourself.

## Rule

Most surprising failures are the agent's own stale state, not the system's behavior. So
you run one cheap ladder of checks **first**, once, before forming any theory — and then
you never run it again on that same unchanged failure. The single pass is the remedy; a
second pass is a substitute for thinking.

## Procedure

Run every rung once, in order. Each rung has a check and a tell — the tell is how you
know whether that rung was the problem.

1. **Unsaved file.** Re-read the file you edited from disk. *Tell:* the edit is missing
   or the content differs from what you believe you wrote.
2. **Stale process.** Identify the running process, server, watcher, or REPL and restart
   it. *Tell:* it was started before your edit, or its log shows no reload.
3. **Stale cache or artifact.** Clear the relevant build, module, bundler, or test cache
   and rebuild. *Tell:* the failure changes or disappears after the clean rebuild.
4. **Wrong branch or worktree.** Check the current branch, the working directory, and
   whether another worktree or checkout of this repo exists. *Tell:* you are not on the
   branch the change landed on, or the running path is not the path you edited.
5. **Wrong file.** Confirm the file being executed is the file you edited — resolved
   import path, installed vs. local copy, `dist/` vs. `src/`, a shadowing duplicate.
   *Tell:* two files match the name, or the import resolves elsewhere.
6. **Outdated generated output.** Compare the timestamp of compiled, generated, or
   transpiled output against its source. *Tell:* the output is older than the source.

7. **The ladder is now spent.** Whatever you found, reconcile it and re-run the failing
   thing exactly once. If it still fails, the cause is real. Stop guessing at state:
   read the actual error, form a theory, and gather evidence for it.

## Evidence

Name the rungs and what each one showed, so nobody — including you — is tempted to climb
it twice:

> Ladder run once: file on disk matches the edit; dev server restarted (was 40 min old);
> `.next/` cleared and rebuilt; on `feat/auth`, single worktree; import resolves to
> `src/auth.ts`, the file I edited; no generated output involved. Failure unchanged, so
> it is not stale state. Diagnosing the 401 now.

"I restarted it and it still fails" is not evidence. It is one rung out of six.

## Boundary

- An **environment** that is unavailable or misconfigured — service down, dependency not
  installed, env var unset, port taken — `ya-comiste`.
- Verifying that a **fix actually worked** — `a-ver-ensename`.
- This skill governs the first sixty seconds of an investigation, and nothing after.

## Exit criteria

You may proceed once all six rungs have been run in this investigation and step 7 has
been executed. From that moment the ladder is closed for this failure.

**Never run the ladder a second time on the same unchanged failure.** No second restart,
no second cache clear, no "let me just try rebuilding again." If the same symptom
survived one honest pass, more state-clearing is avoidance. The only thing that reopens
the ladder is a genuinely new change — you edited something, switched branches, or the
symptom itself changed. Vaporub before bed is fine. Vaporub three nights running means
you should have gone to the doctor.

## Cómo te regaña

> "¿Te duele? Ponte Vaporub."
> "Un tecito y se te quita."
> "Si mañana sigues igual, ya vamos al doctor."

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.