- Home
- Skills
- APIs & Backend
- Dependency Vetting: Check Before Your Agent Installs
Works with the AI Agents you already use
Dependency Vetting: Check Before Your Agent Installs
They run `postinstall` scripts before anyone has reviewed anything.
$10
Dependency Vetting: Check Before Your Agent Installs
Example session with this skill installed
Add YAML config file support to our Python CLI and a nicer colored progress bar for long downloads.
- Read your context and instructions
- Compiled the dependency vetting
Example session (illustrative).
Before installing anything, I vetted two candidate dependencies.
1. YAML parsing
- Necessity: the stdlib has
tomllibbut no YAML parser, so YAML needs a dependency. (If you're open to TOML config instead, we need nothing new.) - Identity:
import yamlis provided by
PyYAML. There is no yaml project on PyPI, so pip install yaml would fail today and is exactly the kind of unclaimed name a squatter could register. The PyPI JSON API shows PyYAML's first upload in 2011 and a source link to github.com/yaml/pyyaml, and the project page lists several maintainers.
- License: MIT. Compatible with our Apache-2.0 project.
- Install code: wheels are available for our Python versions, so the sdist build code is not run.
- Transitive:
pip install --dry-run PyYAMLshows +1 package. - Audit:
pip-audit -r requirements.txtreports no known vulnerabilities. - Decision: Add. I'll use
yaml.safe_loadonly, neveryaml.load.
2. Progress bar. The first candidate a search turned up was richbarx.
- Identity: the package exists, but it was first published 9 days ago by a single new account. Its repository field points at the popular
richproject, whose repo does not publish this name (starjacking pattern). - Decision: Reject. This is a likely lookalike.
- Alternative:
tqdm, which is already in our lockfile as a transitive dependency of huggingface-hub (pip show tqdm, Required-by). Promoting it to a direct dependency adds 0 packages. License: MPL-2.0 and MIT.
Add with caveat: MPL-2.0 is weak copyleft. It's fine as an unmodified dependency, and I've noted it.
Lock: uv add PyYAML tqdm updated uv.lock. The diff touches only those two entries.
PR note:
New dependency: PyYAML@6.x (PyPI)
Why: YAML config parsing; stdlib has no YAML; TOML alternative offered
License: MIT | Install scripts: none (wheel) | Transitive: +1 | pip-audit: clean
Promoted: tqdm (already transitive) | License: MPL-2.0 AND MIT (caveat noted)
Rejected: richbarx (9 days old, lookalike repo link)
Connects securely to your tools. The creator never sees your data.
About this skill
The problem. Coding agents add dependencies the way they autocomplete: fast and without asking. They pip install an import name that belongs to a different package. They invent plausible package names, and attackers register those names in advance ("slopsquatting"). They pull 80 transitive packages for a one-line helper. They run postinstall scripts before anyone has reviewed anything. Licenses never get checked, and the new dependency shows up in a lockfile diff nobody reads.
What the agent does with this skill. Before it runs npm add, pip install, uv add, cargo add, or go get, the agent works through a fixed procedure:
- Necessity: checks the stdlib first, with concrete per-language replacements, then existing direct and transitive dependencies, then whether it could write the code itself in under 30 lines.
- Identity:
- Confirms the package exists and checks its creation date and repository with
npm view, the PyPI JSON API,cargo infoplus the crates.io API, andgo list -m -versions. - Checks for typosquats, scope confusion, PyPI import-name versus install-name traps, dependency confusion, and starjacking.
- Confirms the package exists and checks its creation date and repository with
- Maintenance: looks at release cadence, number of maintainers, recent ownership changes, and very new versions.
- License: reads the actual LICENSE file and applies a compatibility table covering permissive, weak copyleft, strong copyleft, AGPL, and source-available licenses.
- Install-time code: covers npm lifecycle scripts, including the implicit
node-gypbuild, PyPI sdists, Rustbuild.rsand proc-macros, and Go cgo. It evaluates in a scratch directory with scripts disabled. - Transitive weight: compares the dependency tree before and after the add.
- Vulnerabilities: runs
npm audit,pip-audit,cargo audit, andgovulncheck. - Pinning: follows lockfile and pinning rules for each ecosystem.
- Decision: picks add, add with caveat, reject, or ask the user from an explicit decision table, then writes a short justification note for the PR.
Who it's for. Teams and solo developers who let Claude Code, Cursor, Codex CLI, or other agents edit manifests, and who want supply-chain hygiene without reviewing every lockfile by hand.
Requirements. The ecosystem's package manager. The auditor for the language you use (npm audit is built in; pip-audit, cargo-audit, and govulncheck install in one line). Network access to the registry. It works without AgentHarness. If AgentHarness is installed, harness scan --staged is an optional extra check on new dependencies.
Requires network access for registry lookups. User must have ecosystem tools (npm, pip-audit, cargo-audit, etc.) installed for full verification. Cannot vet obfuscated runtime logic.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 17 days ago
Requires
Permissions
Allowed Hosts