Skip to content
Agentmart
Submit a tool

verify-commands

Apply whenever you write or review a command that decides whether work is done: a test invocation, an acceptance check, a CI gate, a shell assertion, a…

Skill by mdsmithaustin

IndexedMIT★ 7
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/mdsmithaustin-verify-commands

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Apply whenever you write or review a command that decides whether work is done: a test invocation, an acceptance check, a CI gate, a shell assertion, a grep-based verify step. Names the failure signal, grounds paths, and catches the shapes that pass green while measuring nothing.

Use cases: testing, devops, data, writing

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md and code. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/verify-commands

What's inside

Skill name
verify-commands
Description
Apply whenever you write or review a command that decides whether work is done: a test invocation, an acceptance check, a CI gate, a shell assertion, a grep-based verify step. Names the failure signal, grounds paths, and catches the shapes that pass green while measuring nothing.
License
MIT
Instructions
31 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# Verify commands

A verify command is an acceptance test. Before shipping one, answer: **if this command were silently doing nothing, what in its output would tell me?** If you cannot answer, you have a command, not a check. Fix the command; do not invent a statement for it.

## Name the failing direction

Next to every runnable check, write one line naming the observable signal that means failure: `non-zero exit`, `"0 passed" in the summary line`, `the coverage line is absent`, `stderr contains "ECONNREFUSED"`. "The command fails" and "an error occurs" are restatements, not signals. `TBD`, `N/A`, and `none` are refusals. Short is fine; `non-zero exit` is complete. Prefer the signal the tool actually emits over one you imagine: for a command that already ran, state what you saw; for a new one, take the signal from the tool's documented output. N commands need N statements.

## Ground every path

Inherit the command that already worked in this repo, verbatim, before deriving a new one. A `cd` target or `--prefix` target must exist, or be created by an earlier step, and must hold the matching manifest (`package.json`, `Makefile`, `pyproject.toml`). Prefer `npm --prefix <dir> run <script>` over `cd <dir> && npm run <script>`; it does not depend on the caller's working directory. Do not hand-reason the filesystem: list it. If you cannot ground a path, say so instead of guessing.

## Shapes that pass while measuring nothing

| Trap | Why it passes | Instead |
|---|---|---|
| `pnpm ls vite \| grep -E '^vite@7'` | tree output starts each line with `└──`, so `^` never matches | drop the anchor, or match `(└\|├)── vite@7` |
| `npm test \| grep '714 passed'` | a count recalled from memory, stale from training | `grep -E '[0-9]+ passed'`, or run the count first |
| `X=$(cmd 2>/dev/null \|\| echo 0)` feeding `[ "$X" = "$Y" ]` | both sides default to `0` when inputs are missing, so parity passes on nothing | let the failure propagate, or guard with `test -f` and exit |
| `grep -c 'LITERAL' file == 0` while an instruction or comment quotes `LITERAL` | the gate trips on comment text, not on code | describe the forbidden thing by concept, not by literal |
| `! grep -q PATTERN file` when a sibling change legitimately adds `PATTERN` elsewhere in the file | a file-wide negative is unsatisfiable across regions | scope it: `awk '/^def f/,/^def /' file \| grep`, or an AST check, or a unit test |
| a test-name filter that matches zero tests | exit 0 with nothing asserted | list tests first (`--collect-only`, `--list`), then run the one named test |
| `npm test \| tail -20` whose exit status is read as the verdict | a pipeline exits with its last command's status, so `tail` reports 0 while `npm test` failed | start the same command line with `set -o pipefail; npm test \| tail -20` (bash, zsh). A `set` from an earlier tool call is gone |

The anchor trap has a family: `npm ls`, `yarn list`, `docker ps`, and `kubectl get` all print column or tree prefixes. `|| echo default` is acceptable only when absence is the default and the value is not used in a comparison meant to detect absence.

## Reviewing someone else's command

Report; never prescribe a replacement path you did not verify. A prescribed path is copied verbatim and carries zero information. A missing directory or manifest is a blocker: quote the raw target. A dynamic path, glob, `~`, or variable is a warning, not a licence to guess the literal.

From mdsmithaustin/pstack, MIT.

More from mdsmithaustin/pstack

Signals

Stars
7
Forks
0
Open issues
19
Last commit (this folder)
Oct 19, 2026 (9 days ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

Same skill elsewhere

Copies of this SKILL.md published by others; search shows only one of them.

More from mdsmithaustin

Cite this page

verify-commands skill. Agent Mart. https://agentmart.dev/skills/mdsmithaustin-verify-commands (last updated 2026-10-24)