# Coverage

> What Keryn checks, what it does not, and how every output states how much of its rule library actually ran on your code.

Source: https://keryn.dev/docs/concepts/coverage/

A clean report means something only if you know what was checked. This page has two halves: what kinds of checks Keryn has, and how each scan states what ran.

## What Keryn covers

| Area | What it checks | How you run it | Guide |
|---|---|---|---|
| **Source code (SAST)** | AI-application and general injection bugs in more than 20 languages, with two engines and a proof level on every finding | `keryn scan` | [Static analysis](https://keryn.dev/docs/guides/sast/) |
| **Dependencies (SCA)** | Locked packages against OSV advisories, decided per advisory range; licences (SPDX) against a deny / allow policy; malicious-package, typosquat, install-script and dependency-confusion signals (npm, PyPI); CycloneDX SBOM intake; reports of other dependency scanners | `keryn import` in 0.5.2; in `keryn scan` and `keryn sca` in the next release | [Software composition analysis](https://keryn.dev/docs/guides/sca/) |
| **Secrets** | Committed credentials, values redacted | In every `keryn scan` in the next release | [Secrets](https://keryn.dev/docs/guides/secrets/) |
| **Agent skills and instruction files** | `SKILL.md`, `CLAUDE.md`, `AGENTS.md` and similar files: hidden instructions, piped remote scripts, unscoped shells | Skill rule packs in `keryn scan` in 0.5.2; a dedicated screen in every scan in the next release | [Agent skill and instruction files](https://keryn.dev/docs/guides/agent-skills/) |
| **Binaries and model files** | What a compiled file is and can do, from its bytes, never executed | `keryn artifact analyze`; Keryn Web triages them inside a folder or zip; `keryn scan` in the next release | [Binary analysis](https://keryn.dev/docs/guides/binary-analysis/) |
| **Agent workflow exports** | n8n, Flowise, Langflow and other exports as graphs: tools, approvals, dangerous compositions | `keryn workflow analyze` | [Agent workflow exports](https://keryn.dev/docs/guides/workflows-and-binaries/) |
| **Running web apps (DAST)** | Captured traffic against passive web rules, scoped ZAP scans, imported Burp or ZAP reports | `keryn dast`, `keryn import` (next release) | [Test a running web app](https://keryn.dev/docs/guides/dast/) |
| **AI security boundaries** | Authenticated replays against a test deployment: tenant isolation, tool authorisation, fix verification | `keryn ai-test` (next release) | [Test AI security boundaries](https://keryn.dev/docs/guides/ai-testing/) |
| **Trust and release decisions** | The edges, chains and chokepoints of an agent, and a Go or Hold answer for a release | `keryn trust-graph`, `keryn decide` | [Trust Canvas](https://keryn.dev/docs/guides/trust-canvas/), [Release decisions](https://keryn.dev/docs/guides/release-decisions/) |
| **Known fixes** | Code that a published advisory's fix removed | `keryn scan` | [Known-fix matching](https://keryn.dev/docs/guides/known-fix/) |

"The next release" means the work is on Keryn's main branch and is not in 0.5.2; each guide says so at the feature.

## What it does not cover

- It does not run your code, except in the optional, isolated [Verify](https://keryn.dev/docs/concepts/limits/) step.
- General (non-AI) vulnerability recall is low. A clean scan is not a statement that code is safe. Measured limits are on [what Keryn does not do](https://keryn.dev/docs/concepts/limits/).
- Languages and file types without a bundled rule get no findings from the library. [Write your own rules](https://keryn.dev/docs/guides/custom-rules/) to add them.
- Nothing here is a compliance determination.

## What a scan states about itself

Every Keryn output states:

- **rules**: how many rules executed, and which were unsupported or only partly supported for the languages present, or timed out,
- **files**: how many files were scanned and how many were skipped (size limits, symlinks, binary files, files your `.gitignore` excludes).

When coverage is incomplete, `keryn scan` exits with **3** unless you pass `--allow-incomplete`. CI never passes on a scan that did not run.

Engines that cannot run say so once, in the report, and the scan goes on without them. That includes Deep in Keryn Web, and the dependency check of an offline scan: a report that did not check its lockfiles says how many, rather than showing an empty list.

The AI-BOM carries the same numbers as `keryn:coverage.rules` and `keryn:coverage.files`, next to its caveat: absence is not evidence of absence, and presence does not prove runtime use.
