Works with the AI Agents you already use
dependency-scan: vulnerable, outdated and unpinned packages
It reports only and edits none of your files; the build tools it runs only with your own yes may write their usual build output.
Free
dependency-scan: vulnerable, outdated and unpinned packages
Example session with this skill installed
scan the dependencies
- Read your context and instructions
- Compiled the dependency-scan
Summary
A real run on a small Python project with a uv.lock, with the network question declined: 2 LOW findings, score 9.8/10. The project, an internal package and its repository are replaced with my-app, internal-lib and github.com/acme/internal-lib; the commit hash and the temp-folder path are replaced with placeholders. Empty sections, the notes block and part of one Coverage row are cut. Counts, scores and wording are as the run printed them.
=== EXECUTIVE SUMMARY ===
dependencies: 0 CRITICAL, 0 HIGH, 0 MEDIUM, 2 LOW
TOTAL ISSUES: 2
not checked offline: Known vulnerabilities (Python), Outdated packages (Python)
=== LOW ===
pypi:internal-lib is a VCS dependency on the branch main
- File: pyproject.toml:7
- Problem: pypi:internal-lib is a VCS dependency on the branch main
- Evidence: pyproject.toml:7: "internal-lib @ git+https://github.com/acme/internal-lib@main"; uv.lock:475 pins commit <commit>
- Impact: The branch moves; the next lock (uv lock --upgrade-package internal-lib or a fresh lock) pulls whatever main holds then
- Fix: Pin the dependency to a tag or a commit hash in pyproject.toml instead of main
- Source: agent-read
- Confidence: high
pypi:pytest-cov is declared in the dev group and nothing uses it
- File: pyproject.toml:25
- Problem: pypi:pytest-cov is declared in the dev group and nothing uses it
- Evidence: pyproject.toml:25 "pytest-cov>=7.1.0" (dev dependency group); no --cov, addopts, pytest config or script anywhere in the repository references it (README tests command is `uv run pytest && uv run ruff check .`)
- Impact: An unused dev package adds install time and attack surface
- Fix: Remove it from the dev group, or add the coverage run that needs it
- Source: agent-read
- Confidence: medium
=== SCORE ===
dependencies: 9.8/10
TOTAL: 9.8/10
=== COVERAGE ===
| Check | Ran with | Result |
|-------|----------|--------|
| Known vulnerabilities (Python) | skipped | not checked offline: user declined network |
| Outdated packages (Python) | skipped | not checked offline: user declined network |
| Unused packages (Python) | agent-read | 1 finding |
| Pins and lockfile (Python) | agent-read | 1 finding (uv.lock present and committed; 0 drift) |
| Scanner (pip-audit) | skipped | not run: user declined network; vulnerabilities not checked |
| Scanner (osv-scanner) | skipped | absent: manifests and uv.lock read for pins |
| Non-production paths | agent-read | 0 findings dropped |
| Findings file | command:write | <temp folder>/dependency-scan-<UTC timestamp>.jsonl |
Scanners present: pip-audit, uv; absent: osv-scanner, trivy. Reading finds less than a scanner
would; to close the gap on vulnerabilities, allow the network so pip-audit can run on the
exported uv.lock. No scanner contacted a database or registry in this run.
=== ACTION PLAN ===
1. CRITICAL: none
2. HIGH: none
3. MEDIUM: none
4. LOW: pin internal-lib to a tag or commit (pyproject.toml:7); remove or use pytest-cov (pyproject.toml:25)
Next steps
The vulnerability and outdated checks show as "not checked offline" because that run declined the network question. Run again with --allow-network so pip-audit can check the lockfile. The score of 9.8 covers only what was checked; it is not a statement that the dependencies are free of known vulnerabilities.
Connects securely to your tools. The creator never sees your data.
About this skill
dependency-scan is one skill that checks the packages a project declares and locks. Ask "scan the dependencies", "any vulnerable packages?" or "do we have a lockfile?" and it finds the manifests at any depth, treats each folder as one project, runs the checks for each ecosystem and prints one report with a score, a Coverage table and an action plan. It reports only and edits none of your files; the build tools it runs only with your own yes may write their usual build output.
| Ecosystem | Manifests |
|---|---|
| Python | requirements*.txt, pyproject.toml, Pipfile, setup.py, setup.cfg |
| npm | package.json |
| Go | go.mod |
| Rust | Cargo.toml |
| Java | pom.xml, build.gradle, build.gradle.kts |
| Ruby | Gemfile, *.gemspec |
| PHP | composer.json |
| .NET | *.csproj, *.fsproj, packages.config, Directory.Packages.props |
Six rules
| Rule | Severity | What it is |
|---|---|---|
deps-known-vuln | The scanner's rating (MEDIUM when it gives none) | A declared or locked version with a published advisory. Needs a scanner |
deps-outdated | LOW; MEDIUM when deprecated, yanked or end-of-life | A direct dependency behind the latest release |
deps-unused | LOW | A declared direct dependency that nothing imports |
deps-unpinned | MEDIUM | A floating or range specifier that nothing locks |
deps-no-lockfile | MEDIUM for an application, LOW for a library | A project with no lockfile |
deps-lockfile-drift | MEDIUM | A dependency missing from the lockfile or locked at a version the manifest does not allow |
What makes it different
- Evidence on every finding. Each finding has a file and line, or the scanner output it came from, plus its source (
scanner:<name>oragent-read) and a confidence level. - No made-up vulnerabilities. An advisory id (CVE, GHSA, PYSEC, RUSTSEC) or a fixed version appears only when a scanner printed it. With no scanner and no network, the check is marked "not checked offline", never shown as clean.
- A Coverage table. The report says which checks ran with which tool, which were skipped and why, and which scanners were present or absent.
- Works without scanners. With none installed, it reads manifests and lockfiles for pins, unused packages and drift, and marks those findings
agent-read. With pip-audit, npm audit, osv-scanner, cargo audit, composer audit, govulncheck or the like installed, it uses them. It never installs one; it prints the install line for you to run. - Monorepos and workspaces. Each manifest folder is its own project, and workspace members use the workspace root's lockfile (npm, yarn and pnpm workspaces, Cargo, uv, Go).
- One question about the network. Lookups that query a vulnerability database or a registry need one yes per scan.
--allow-networkgives it in advance. Without it, those checks are skipped and said so. - Safe by default. Report only. The one file it writes itself in your repository is the export, and only with
--export-report. Project-code scanners (Maven, Gradle,dotnetrestore) run only after their own yes and may write build output.
Flags
--export-report saves the report to docs/audit/DEPENDENCY_SCAN_REPORT_<YYYY-MM-DD>.md. --allow-network says yes to the network question in advance. --only dependencies.
Known limitations
Known-vulnerability coverage depends on the scanner: with none installed and no network, those checks are reported as "not checked offline", not as clean. Offline, the outdated check does not run. Unused, unpinned, lockfile and drift checks work from the files alone, with lower confidence when read by hand. Manifests under test/, tests/, spec/, __tests__/, fixtures/, examples/, vendor/ and node_modules/, and *.example files, are left out and counted in Coverage. It does not check your own code, secrets, Dockerfiles, pipelines or the licences of dependencies. Findings from an agent's reading should be checked before you act.
Install
Claude Code. Unzip it into ~/.claude/skills/ as one folder, dependency-scan/, and keep the folder whole.
Want fixes, CI and run memory?
This scan only reports. The Full Audit Pack runs the same dependency checks and adds: updates by patch then minor that revert when your tests break (--apply), a CI mode with exit codes, run memory that shows what is new, recurring or fixed, and seven more audits in one report and one score: https://agensi.io/skills/full-audit-pack-8-code-audits-one-score-memory
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Security.
Known-vulnerability coverage depends on the scanner: with none installed and no network, those checks are reported as "not checked offline", not as clean. Offline, the outdated check does not run. Unused, unpinned, lockfile and drift checks work from the files alone, with lower confidence when read by hand. Test, fixture, example, vendor and node_modules folders are left out and counted in Coverage. It does not check your own code, secrets, Dockerfiles, pipelines or licences. Findings from an agent's reading should be checked before you act.
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 buying a skill.
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
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 1 day ago
- Free with an Agensi account
- Software
Pythondepcheck
Node.js
- Optional
- depcheck, for unused Node packages
No scanner is required. With pip-audit, npm audit, osv-scanner, govulncheck, cargo audit or similar installed, it checks known vulnerabilities; without one, those checks are reported as not checked, never as clean.
Permissions
File Scopes
Terminal: runs the scanners you already have installed. Read Files: reads manifests and lockfiles. Write Files: only the report, with --export-report (docs/audit/), plus a findings file in your temp folder. Network: none by default; after your yes, OSV (api.osv.dev), the GitHub Advisory Database, vuln.go.dev, the NVD (OWASP dependency-check) and the package registries your scanners query (PyPI, npm, crates.io, Go proxy, RubyGems, Packagist, NuGet, Maven Central). Only package names and versions are sent, never source code.