- Home
- Skills
- Legal & Compliance
- Dependency License Audit: Flag AGPL and unknown licenses
More screenshots
Works with the AI Agents you already use
Dependency License Audit: Flag AGPL and unknown licenses
Find the AGPL package before your lawyer does. Classify every dependency license in a package-lock as allow, review, deny or unknown against your own policy, with an exit code for CI.
$12
Dependency License Audit: Flag AGPL and unknown licenses
Example session with this skill installed
Audit references/demo.json using the built-in example policy. Return Markdown and prioritize release blockers.
- Read your context and instructions
- Compiled the dependency license audit
Dependency license audit
Total: 5 | Allow: 1 | Deny: 1 | Review: 1 | Unknown: 2
| Dependency | Version | Normalized license | Status | Categories |
|---|---|---|---|---|
| example-safe | 1.0 | MIT | allow | other/unclassified |
| example-server | 2.0 | AGPL-3.0-only | deny | strong-copyleft |
| example-ambiguous | 1.2 | BSD | unknown | other/unclassified |
| example-library | 3.0 | LGPL-2.1-only | review | weak-copyleft |
| example-missing | 0.1 | (missing) | unknown | unknown |
Notices
- Not legal advice. Metadata is unverified; policy approval is not legal compatibility.
- OR permits an allowed alternative; record and comply with the chosen license before distribution.
- Review is advisory (exit 0 unless deny/unknown exists); unknown terms always fail, even in OR.
- Copyleft categories describe all branches; exceptions may change obligations. Inspect license texts.
- Retain required copyright, license and NOTICE texts; this report does not generate legal notices.
Blocked under the example policy (exit 1). Deal with example-server first. Then get the exact BSD variant for example-ambiguous and any license evidence for example-missing, and send example-library to review. Check your organization's own policy before treating this as a release decision.
Connects securely to your tools. The creator never sees your data.
About this skill
The problem
You're about to ship, and the lockfile lists hundreds of dependencies. You need to know which licenses your policy allows, which it forbids, and which packages don't declare a license at all. Scrolling through package-lock.json by hand doesn't scale, and a green CI run can hide licenses that still need a reviewer's sign-off.
See it before you buy: real input, real output Command (from the package folder, exit code 1, 0.026 s):
python3 -I -B scripts/license_audit.py references/demo.json --format markdown
Before: Dependency list (excerpt of references/demo.json)
"name": "example-server",
"license": "AGPL-3.0-only"
{
"license": "BSD"
{
"license": "LGPL-2.1-only"
{
"name": "example-missing",
"version": "0.1"
After: lines copied from the script's output (excerpt):
Total: 5 | Allow: 1 | Deny: 1 | Review: 1 | Unknown: 2
| example-server | 2.0 | AGPL-3.0-only | deny | strong-copyleft |
| example-ambiguous | 1.2 | BSD | unknown | other/unclassified |
| example-library | 3.0 | LGPL-2.1-only | review | weak-copyleft |
| example-missing | 0.1 | (missing) | unknown | unknown |
What the numbers show
- Ran on 9 public package-lock files (2,898 dependency entries): 1,667 allowed, 0 denied, 53 needing review and 1,178 unknown. 8 of the 9 produced a report.
- The ninth stopped with exit 2 because one package declares its license as the phrase 'Public Domain', which isn't a valid license expression.
- Median runs were from 26.37 to 37.57 ms, Python startup included.
What you get
- One Python script with no dependencies, plus instructions that tell your agent how to run it and explain the results.
- It reads JSON inventories, CSV files, package-lock v1, v2 and v3, and requirements.txt (with a license map you write, since requirements files don't carry licenses).
- Every dependency gets one of four results: allow, deny, review or unknown. You can use your own policy file. If you don't, the agent uses a built-in example policy and says so.
- It understands expressions like
MIT OR Apache-2.0andGPL-2.0-only WITH Classpath-exception-2.0. It knows 40 common license IDs, five loose spellings such as "MIT License" and "Apache 2.0", and four exceptions. Anything outside that list is marked unknown rather than guessed. - Output is JSON for scripts or Markdown for people. The agent leads with the blockers, then lists what evidence to collect and which items need a reviewer.
- Exit codes you can gate on: 1 if anything is denied or unknown, 2 if the input is broken, 0 otherwise. Review items alone don't fail the build.
Measured on real data (full detail) I ran it on nine package-lock files from public projects, five times each. Eight produced a report. Together they cover 2,898 dependency entries: 1,667 allowed, 0 denied, 53 needing review and 1,178 unknown. Of the unknowns, 1,092 had no license field at all and 86 were BlueOak-1.0.0, which the script doesn't recognize. The same package installed in two places counts twice. Times are wall-clock per run, Python startup included.
| Input | Entries | Review | Unknown | Median ms | Exit |
|---|---|---|---|---|---|
| axios/docs/package-lock.json | No report | — | — | 29.96 | 2 |
| axios/package-lock.json | 678 | 14 | 4 | 37.57 | 1 |
| axios/tests/module/cjs/package-lock.json | 83 | 1 | 0 | 27.09 | 0 |
| axios/tests/module/esm/package-lock.json | 72 | 12 | 0 | 26.48 | 0 |
| axios/tests/smoke/cjs/package-lock.json | 100 | 1 | 4 | 27.34 | 1 |
| axios/tests/smoke/esm/package-lock.json | 69 | 12 | 0 | 26.37 | 0 |
| dotenv/package-lock.json | 738 | 1 | 394 | 35.19 | 1 |
| node-express-realworld-example-app/package-lock.json | 773 | 0 | 773 | 30.26 | 1 |
| typescript-starter/package-lock.json | 385 | 12 | 3 | 31.54 | 1 |
The axios/docs file stopped with exit 2 because one package declares its license as the phrase Public Domain, which isn't a valid license expression. In an earlier pass, 30 findings (16 review, 14 unknown) were hand-checked against the lockfile records and the policy, and all 30 were classified correctly. This rerun reproduced the same counts. That confirms the script reads the metadata and applies the policy correctly. It says nothing about whether the declared licenses are true, or about legal risk. Commands, per-run timings and the hand-check table are in references/benchmarks.md.
Known limitations
- It only reads the license fields you give it. It doesn't scan source code, query npm or PyPI, install anything, or check license texts.
- It knows 40 license IDs. Valid SPDX IDs outside that list come back unknown, and a malformed expression stops the whole run with exit 2.
- Review items don't fail the exit code. If your process needs sign-off, check
summary.reviewyourself. - It counts every lockfile entry, dev and optional ones included. There's no production-only filter. The whole input is loaded into memory.
- "Allowed" means it passed your policy, not that it's legally compatible. It doesn't generate an attribution or NOTICE file.
Who it's for Maintainers and release reviewers who already have a lockfile or inventory and want the same policy check every time, with a clear list of what to fix next.
Requires Python 3.8 or newer and the input files on your machine. Requirements mode also needs a license map you provide. No extra libraries and no account. It runs locally, makes no network calls, and writes nothing unless you redirect the output.
Check it yourself
30 command-line test cases and four saved sample reports, covering every exit code, all three lockfile versions, license expressions and broken inputs. To reproduce the demo, run python3 -I -B scripts/license_audit.py references/demo.json --format markdown from the package folder (exit 1). For a passing JSON report, use references/passing.json instead. The test cases and expected results are in references/checks.json. Separately, nine public lockfiles were run five times each: eight finished, and one was rejected for a non-standard license phrase.
Price and guarantee $12 one time, yours to use with any agent that can run Python. Covered by Agensi's money-back guarantee if it doesn't do what this page promises.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Legal & Compliance.
Bounded to 40 license IDs. No source scanning or network lookups. Supports UTF-8 only. Requirements mode has limited syntax support (no extras/markers). Not legal advice.
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
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 15 days ago
- Software
Python 3
Permissions
Uses a terminal and file-reading tools to read the supplied local inventory, optional policy and optional license map. Reports go to stdout and errors to stderr; nothing is written unless the user redirects output. No network access. Nothing is sent anywhere; no telemetry or subprocesses in the shipped script.