@quantakrypto/qscan

@quantakrypto/qscan

Find quantum-vulnerable cryptography in any codebase.

qscan walks a project and reports where it relies on classical asymmetric cryptography (RSA, (EC)DH, ECDSA, EdDSA, DSA, …) — the algorithms broken by a sufficiently large quantum computer, and the ones exposed to harvest-now, decrypt-later (HNDL) attacks today. It scans source files, dependency manifests, and configuration, then prints a readiness score and a concrete next step.

• Zero runtime dependencies. Node built-ins only; the engine is @quantakrypto/core.
• CI-friendly. Severity thresholds drive the exit code; baselines suppress known findings so the build only fails on new problems.
• Multiple formats. human (default), json, SARIF 2.1.0 for code-scanning dashboards, and a CycloneDX 1.6 CBOM for compliance tooling.
• Fast on big repos. Optional worker-thread parallelism (--parallel) and git-aware incremental scanning (--changed).

Language coverage

Inline crypto detection currently covers JavaScript/TypeScript, Python, Go, Java/Kotlin/Scala (JCA + BouncyCastle), C#/.NET, Rust, Ruby, PHP (openssl / phpseclib3 / libsodium), Elixir (:crypto / X509 / JOSE), and C/C++ (OpenSSL, Mbed TLS, wolfSSL) source. PEM key material, SSH keys, TLS/certificate config, and dependency manifests for six ecosystems — npm (plus yarn.lock / pnpm-lock.yaml), PyPI, Cargo, Go modules, Maven, and RubyGems — are detected in any file regardless of language.

qScan is honest about coverage: if a scan walks files but finds none in a supported source language, it says so and will not present a bare 100/100 as a clean bill of health — the crypto may simply live in a language it can't read yet. Always check the analyzed count in the report before relying on the score.

Install

npm install -g @quantakrypto/qscan
# or run without installing
npx @quantakrypto/qscan .

Requires Node ≥ 20.

Usage

qscan [path] [options]

path defaults to the current directory (.).

Options

Flag	Description	Default
--format <human|json|sarif|cbom>	Output format.	human
--cbom	Alias for --format cbom (CycloneDX 1.6 CBOM).	—
-o, --output <file>	Write the report to a file instead of stdout.	stdout
--severity-threshold <level>	Exit 1 if any finding is at/above this level. One of critical, high, medium, low, info.	high
--no-source	Skip scanning source files for inline crypto.	scan on
--no-deps	Skip scanning dependency manifests.	scan on
--no-config	Skip scanning config files (TLS/certificates). Toggles config detectors — not the config file below.	scan on
--config <path>	Use this quantakrypto.config.json instead of auto-discovering one at the scan root.	auto
--no-config-file	Disable quantakrypto.config.json auto-discovery.	discovery on
--ignore <pattern>	Exclude paths matching <pattern>. Repeatable.	—
--include <pattern>	Restrict the scan to paths matching <pattern>. Repeatable.	all files
--max-file-size <bytes>	Skip files larger than <bytes>.	2 MiB
--no-default-ignores	Don't skip node_modules/.git/dist by default.	ignores on
--scan-minified	Scan minified/generated/bundled files too.	skipped
--changed	Incremental: scan only files git reports as changed.	off
--since <git-ref>	With --changed, diff against <git-ref> (implies --changed).	working tree
--parallel	Scan using a worker-thread pool when the workload is large enough.	off
--concurrency <n>	Worker count for --parallel (implies --parallel). 0/1 forces serial.	CPU count
--triage	BYOK LLM pass that re-ranks findings by real exposure and explains them. Never suppresses; never changes the exit code. Needs an API key.	off
--triage-floor <level>	With --triage, only triage findings at/above this level.	medium
--context <level>	How much source is shared with the LLM: metadata, snippet, function, file (secrets always redacted).	snippet
--dry-run	With --triage, print the exact payload that would be sent and exit without contacting the provider.	off
--llm-provider <name>	BYOK provider: anthropic or openai-compatible.	anthropic
--llm-model <id>	Model id for the BYOK provider.	provider default
--baseline <file>	Suppress findings whose fingerprint is in the baseline file.	—
--write-baseline <file>	Write current findings as a baseline, then exit 0.	—
--quiet	Suppress the human summary banner.	off
-v, --version	Print version and exit.	—
-h, --help	Print help and exit.	—

Exit codes

Code	Meaning
0	No findings at/above the threshold — or a baseline was written.
1	One or more findings at/above the severity threshold.
2	Usage error or I/O failure.

Config file (quantakrypto.config.json)

qScan auto-discovers a quantakrypto.config.json at the scan root and applies it under a strict precedence: CLI flags > config file > built-in defaults (per-key). It encodes include / exclude / maxFileSize / noDefaultIgnores / scanMinified, the detector-family toggles, severityThreshold, and a baseline path so CI and local runs share one policy. Point at a file elsewhere with --config <path>, or disable discovery with --no-config-file. See docs/CONFIG.md for the full schema.

Example output

qScan — quantum-vulnerable cryptography report
root: ./examples/vulnerable-app  •  files scanned: 2  •  qscan v0.4.2

3 findings  (2 high, 1 medium)
2 exposed to harvest-now-decrypt-later (HNDL).
Readiness score: 70/100

Top findings
  high     rsa-keygen      src/crypto.js:5
           RSA is not quantum-safe.
           → Use ML-KEM-768 (hybrid X25519MLKEM768).
  high     dep-vulnerable  package.json:7
           Dependency "node-forge" provides classical asymmetric crypto.
  medium   ecdh-usage      src/crypto.js:13
           ECDH is not quantum-safe.
           → Use a hybrid KEM (X25519MLKEM768).

Next step: migrate src/crypto.js — Use ML-KEM-768 (hybrid X25519MLKEM768).

Color is emitted only when writing the human format to an interactive terminal (and is suppressed by the NO_COLOR environment variable). Reports written to a file or piped are always plain text.

Baselines

A baseline records the fingerprints of findings you have already triaged. qScan uses the canonical baseline shared across the whole monorepo (@quantakrypto/core, the GitHub Action, and this CLI): a single on-disk format and a single fingerprint algorithm, so a baseline written by one tool is understood by the others.

A fingerprint is a full SHA-256 of ruleId|file|normalizedSnippet. It is deliberately line-insensitive — unrelated edits that shift line numbers no longer invalidate it — and the snippet's whitespace is normalized so reformatting doesn't either. It ignores volatile fields like severity wording or timestamps.

# 1. Accept the current state of the world.
qscan . --write-baseline qscan-baseline.json

# 2. From now on, fail only on findings that are NOT in the baseline.
qscan . --baseline qscan-baseline.json

The baseline file is plain JSON:

{
  "version": 1,
  "fingerprints": [
    "0f1e2d3c4b5a...full-sha256...",
    "a1b2c3d4e5f6...full-sha256..."
  ]
}

Incremental scans

In CI you usually only care about what a change introduced. --changed restricts the scan to the files git reports as modified, which is a large win on big repos:

# Files modified in the working tree (staged, unstaged, and untracked).
qscan . --changed

# Everything that changed since a base ref (e.g. a PR base).
qscan . --changed --since origin/main

Outside a git work tree the changed-file list is empty (nothing is scanned), rather than an error.

Parallel scans

For large trees, route the scan through a worker-thread pool. qScan automatically stays serial for small inputs, so --parallel is safe to leave on:

qscan . --parallel               # auto worker count (CPU cores)
qscan . --concurrency 4          # pin the worker count (implies --parallel)

CBOM (CycloneDX)

Emit a CycloneDX 1.6 cryptographic bill of materials — one cryptographic-asset component per distinct (algorithm, primitive) pair, with file:line occurrence evidence — for compliance and supply-chain tooling:

qscan . --cbom -o qscan-cbom.json
# equivalently:
qscan . --format cbom -o qscan-cbom.json

The output is deterministic (sorted components and occurrences, stable serial number), so re-running on an unchanged tree produces byte-identical CBOMs.

Triage (opt-in, BYOK)

--triage adds an optional LLM pass that re-ranks and explains findings by their real-world exposure. It is purely additive: it never suppresses a finding and never changes the exit code (that is still computed from severity alone), so it cannot mask a problem or gate CI. It is bring-your-own-key — provide an API key via QK_LLM_API_KEY (or ANTHROPIC_API_KEY / OPENAI_API_KEY); without a key, triage is skipped and the deterministic scan is unaffected.

# Re-rank + annotate findings at/above the floor (default floor: medium).
qscan . --triage

# Only triage the most serious findings, with a specific provider/model.
qscan . --triage --triage-floor high --llm-provider anthropic --llm-model claude-sonnet-5

# See exactly what would be sent — contacts nothing, needs no key.
qscan . --triage --dry-run

--context controls how much source leaves the machine (metadata | snippet | function | file, default snippet); secrets are always redacted. --dry-run prints the exact, redacted payload and exits without contacting the provider, so you can review what triage would send before enabling it.

Remediation (qremediate)

This package also ships a qremediate bin that applies fixes, not just reports them. It scans, then for each finding a deterministic codemod proposes a patch; every patch must clear a verify gate (the crypto finding is gone and no new crypto finding is introduced) and a patch policy (only files that have findings, plus dependency manifests) before it is offered.

qremediate .                 # print a unified diff of every verified fix (default; writes nothing)
qremediate . --mode apply    # write the verified fixes into the working tree
qremediate . --mode pr       # commit them to a new branch and open a DRAFT PR (never merges)

Deterministic codemod fixes are safe to apply directly. With --llm (BYOK, same keys as --triage), an LLM can also propose fixes codemods can't:

qremediate . --llm --mode diff   # review LLM-proposed fixes as a diff first
qremediate . --llm --mode pr     # or stage them as a draft PR

Review LLM-proposed fixes before trusting them. The verify gate only confirms the classical crypto finding was removed (a crypto-count check) — it is not a full semantic safety review, so read every LLM-proposed change as a diff. Prefer --mode diff or --mode pr for --llm fixes; --mode apply writes straight to the working tree. Draft PRs are built in an isolated git worktree (your checkout is never touched) and are never auto-merged.

CI

Fail the build on any high-or-worse finding, while tolerating an accepted baseline:

# .github/workflows/qscan.yml
name: qscan
on: [push, pull_request]

jobs:
  quantum-readiness:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: 20
      - name: Scan for quantum-vulnerable crypto
        run: npx @quantakrypto/qscan . --severity-threshold high --baseline qscan-baseline.json

      # Optional: upload SARIF to GitHub code scanning.
      - name: Generate SARIF
        if: always()
        run: npx @quantakrypto/qscan . --format sarif -o qscan.sarif
      - uses: github/codeql-action/upload-sarif@v3
        if: always()
        with:
          sarif_file: qscan.sarif

Programmatic API

The CLI is a thin shell over runQscan, which @quantakrypto/action reuses:

import { runQscan, EXIT } from "@quantakrypto/qscan";

const { result, exitCode, suppressed } = await runQscan({
  path: "src",
  format: "json",
  severityThreshold: "high",
  baseline: "qscan-baseline.json",
});

console.log(`${result.findings.length} findings, exit ${exitCode}`);
if (exitCode === EXIT.FINDINGS) process.exitCode = 1;

runQscan never touches process or stdout — it returns the rendered report string and a suggested exitCode, leaving I/O to the caller. The package also re-exports the argument parser (parseArgs, defaultOptions), severity helpers (severityRank, meetsThreshold), and the canonical baseline utilities from @quantakrypto/core (fingerprintFinding, baselineFromFindings, applyBaseline, loadBaseline, saveBaseline) plus their legacy aliases (fingerprint, buildBaseline, readBaseline, writeBaseline) for source compatibility.

Examples

See examples/ for a sample vulnerable project and the commands to scan it.

License

Apache-2.0

Support & training

Questions, commercial support, or post-quantum readiness training for your team — visit quantakrypto.com or email [e-mail].