Skip to content
Skills Library
Security · Automation

setup

/setup

Run zero-shelter on this project for the first time and wire it in. Use when someone asks to check dependency vulnerabilities, cut down scanner noise, or install zero-shelter. Korean requests look like: 의존성 취약점 점검해줘, 보안 스캔 돌려줘, zero-shelter 설치.

Licence
Apache-2.0
Last repository change
12.10.2026

Assessment and security

84/ 100Class: good
Format9 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content35 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
1
Scripts
0
HTML comments
0
URLs in SKILL.md
8
Deprecation

No signals detected. Repository changed on 12.10.2026, without references to retired models or APIs.

Licence

Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.

What it does and when to use it

Run zero-shelter on this project for the first time and wire it in. Use when someone asks to check dependency vulnerabilities, cut down scanner noise, or install zero-shelter. Korean requests look like: 의존성 취약점 점검해줘, 보안 스캔 돌려줘, zero-shelter 설치.

Original author description

Run zero-shelter on this project for the first time and wire it in. Use when someone asks to check dependency vulnerabilities, cut down scanner noise, or install zero-shelter. Korean requests look like: 의존성 취약점 점검해줘, 보안 스캔 돌려줘, zero-shelter 설치.

How it works

  1. The second scanner
  2. What else to look for
  3. Recording the backlog
  4. After someone fixes something
  5. A page for a human
  6. Keeping a history
  7. Wiring it in
  8. What this skill must not do
Original SKILL.md headings · 195 lines, 1,247 words
SKILL.md preview
---
description: Run zero-shelter on this project for the first time and wire it in. Use when someone asks to check dependency vulnerabilities, cut down scanner noise, or install zero-shelter. Korean requests look like: 의존성 취약점 점검해줘, 보안 스캔 돌려줘, zero-shelter 설치.
---

# First run

Two things, and the second one is not optional.

```bash
npx --yes zero-shelter judge          # no install needed for this one
```

Read the exit code — it is the answer, not decoration:

| Exit | Meaning | What to do |
|---|---|---|
| 0 | Scanned, nothing new to fix | Say so and stop |
| 1 | New findings | Walk through them (see `/zero-shelter:explain`) |
| 2 | **Could not judge** | Never report this as clean. The message says why — usually no usable lockfile, so create one with the project's package manager first |

Requires Node 20+. If it prints a Node version message, that is the whole
problem; do not try to work around it.

Scope: this skill covers dependency vulnerabilities only (`npm audit` and
`osv-scanner`). Dockerfiles, Terraform/IaC, Actions workflows, secrets, and
first-party source are out of scope — a clean dependency run does not mean
the project as a whole is clean.

## The second scanner

This tool reconciles what two scanners each called by a different name. With one
source there is nothing to reconcile, so **install `osv-scanner` before judging
the result of a first run.** The pinned captures show what the second source
buys beyond a shorter list:

| project | fixed versions: npm audit → both | transitive advice: npm audit → both |
|---|---:|---:|
| juice-shop | 15/73 → 50/82 | 0 → 11 |
| nodegoat | 7/177 → 109/173 | 0 → 49 |
| dvna | 20/51 → 40/51 | 0 → 6 |
| hackathon-starter | 0/11 → 4/11 | 0 → 2 |

The npm-audit-only transitive advice block is empty in every pinned capture; the
table's both-sources column shows what changes when osv-scanner contributes. The
benchmark is reproducible from `bench/captures/` with
`npm run build && node bench/evaluate.mjs`; these figures are evidence about
this project's output, not a promise that every repository has the same ratio.

The noise reduction still helps readability. On uptime-kuma, npm audit alone
reports 71 and leaves 71; add osv-scanner and it is 142 in, 71 out.

```bash
brew install osv-scanner        # macOS / Linuxbrew
go install github.com/google/osv-scanner/v2/cmd/osv-scanner@latest
```

Neither works everywhere. If both fail, the releases page has prebuilt binaries,
and that is a better answer than proceeding with one source and explaining a 0%.

**Then check it actually ran**, rather than assuming the install worked:

```bash
npx --yes zero-shelter judge | grep -q "one source" && echo "STILL ONE SOURCE"
```

A run with one source ends its summary with `one source, nothing to reconcile`.
If that phrase is on screen the second scanner is not contributing, whatever
`brew` printed.

Do not grep the JSON for `osv-scanner`. It appears in `skipped` when the
scanner is **missing**, so the count is non-zero in exactly the case the check
is supposed to catch — an earlier version of this skill said to do that and had
it backwards.

Do not describe a one-source run as a normal result. It is a valid way to run
this and the ranking and baseline still work, but the deduplication this tool
exists for is switched off, and a reader who is not told that will conclude the
tool does nothing.

## What else to look for

`ls` the tree before concluding anything. When several of these are present,
start with Trivy — one install covers several of them.

| You see | Domain | Install |
|---|---|---|
| `Dockerfile` | container | [Trivy](https://github.com/aquasecurity/trivy) (Apache-2.0) — `brew install trivy` |
| `*.tf` | IaC | [Trivy](https://github.com/aquasecurity/trivy) (Apache-2.0); [Checkov](https://github.com/bridgecrewio/checkov) (Apache-2.0) also reads Terraform |
| `.github/workflows/` | CI/CD | [zizmor](https://github.com/zizmorcore/zizmor) (MIT) |
| source with no usable lockfile | first-party code | [Opengrep](https://github.com/opengrep/opengrep) (LGPL-2.1) |
| any history at all | secrets | [Gitleaks](https://github.com/gitleaks/gitleaks) (MIT) |

Why Opengrep and not Semgrep: Semgrep's engine remains LGPL-2.1, but its
rules repository moved to a separate licence restricting some commercial,
SaaS, and competing-product use
([background](https://socket.dev/blog/opengrep-forks-semgrep)), so naming
"semgrep" is itself a licence decision — Opengrep is the consortium-governed
fork created to preserve an open SAST ecosystem.

On secrets: [TruffleHog](https://github.com/trufflesecurity/trufflehog)
(AGPL-3.0) is also a good scanner, but that licence matters for commercial
products, so the default here is Gitleaks (MIT). This skill names tools only;
it does not read their output — ingesting a new scanner's report is separate
work, not part of a first run.

## Recording the backlog

A project that has never run this has a backlog it inherited. Fixing all of it
today is not the goal, and failing CI on it teaches people to switch the gate
off.

```bash
npx --yes zero-shelter judge --update-baseline
```

From then on only new findings are reported and CI fails on the regression this
change introduced. Explain that trade before running it: **anything recorded is
no longer shown**, so run it when the current list has actually been looked at,
not to make output disappear.

## After someone fixes something

Re-run `judge`. Findings that were accepted and are no longer reported get their
own line, so the work that was just done is visible instead of showing up as a
number quietly getting smaller. If a scanner that contributed to the baseline
did not run this time, that line says so — do not upgrade "no longer reported"
into "fixed" when the CLI itself is hedging.

Prune the baseline afterwards with `--update-baseline` so it stops listing
fingerprints nothing produces.

## A page for a human

```bash
npx --yes zero-shelter judge --format html --output zero-shelter.html
```

One self-contained file: the commands first, then every finding with the score
that put it there. Offer it when someone wants to look for themselves, share a
state with a teammate, or read it in Korean (`--lang ko`). It needs no network
and no server; opening the file is enough.

## Keeping a history

```bash
npx --yes zero-shelter judge --record
npx --yes zero-shelter history
```

`--record` appends one line per run to `.zero-shelter/history.jsonl`; nothing is
recorded unless asked. `history` shows what appeared and what stopped being
reported between runs, and the html report grows a section once two runs exist.

Suggest `--record` when a project is going to be judged repeatedly — in CI, or
alongside a baseline. Do not turn it on silently: it writes a file into their
repository, and that is their decision.

## Wiring it in

Offer these; do not add them unasked.

**CI** — append to an existing workflow:

```yaml
- run: npx zero-shelter judge --format sarif --output zero-shelter.sarif
  continue-on-error: true

- uses: github/codeql-action/upload-sarif@6f5948dfacef28e207b48d0905cf90c03365536d # v3.37.9
  with:
    sarif_file: zero-shelter.sarif
```

**Coding agent context** — `.claude/settings.json`:

```json
{
  "hooks": {
    "UserPromptSubmit": [
      { "hooks": [{ "type": "command", "command": "npx zero-shelter hook" }] }
    ]
  }
}
```

The hook puts the current findings into the session so an agent does not add a
dependency this project already has an unfixed advisory for. It never blocks a
prompt and never fails.

## What this skill must not do

- **Do not re-rank, re-judge, or filter the findings.** The ordering is
  computed by the CLI and is reproducible; anything you add on top is not, and
  the entire point of this tool is that its judgement can be checked.
- **Do not report a 0% reduction as a result.** It means one source ran. Say
  which one is missing and offer to install it.
- Do not run `--update-baseline` without saying what it hides.
- Do not describe a run that exited 2 as passing.

Versions

RepositoryScoreLicence
This page · zero-shelter/zero-shelter84Apache-2.0
HwangSlater/harness-fit100MIT
nesquikm/dev-process-toolkit100MIT
userKael/rastro100MIT
als-apg/osprey100BSD-3-Clause
dedujeong/gemba100Apache-2.0
eigenwise/eigenwise-toolshed100MIT
awrshift/agent-memory-kit100MIT
lifecycle-innovations-limited/claude-ops100MIT

Database snapshot: 13.11.2026 · package version 9b2a60012b7c (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 84ci
    Wire zero-shelter into a project's CI so new dependency findings fail the build, without the inherited backlog failing it from day one. Use when someone asks to add a dependency or security gate to CI, gate pull requests on vulnerabilities, upload SARIF to the Security tab, or…
  • 84fix
    Apply the upgrades zero-shelter asked for and check they worked, including transitive findings a top-level install cannot reach. Use when someone wants dependency findings fixed, asks to resolve vulnerabilities, or says to do what the report says. Korean requests look like: 의존성…
  • 100security-feed-check
    Check the pwn.watch and Cloudflare Radar Zero Wire vulnerability feeds against what is actually installed on this machine and in your deployed projects, apply narrow verified local security updates, and report only what affects you with exact fixes. Use when the user asks for a…

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.