Skip to content

    Type a page, a command such as keryn scan, or a rule id.

    Glossary

    View as Markdown

    Keryn uses four sets of words, and they never share a meaning. A decision says what to do with the release. An analysis state says how much was looked at. Severity says how bad a finding is if it is real. Evidence says how much Keryn established about it. Keryn Web, the CLI’s text output, the editors and these docs use the same words; each word comes with an icon, so colour is never the only cue.

    The words in brackets are the machine values. They are what JSON, SARIF, exports and exit codes carry, and they do not change.

    What to do with the release. A decision comes from your policy, applied to what the analysis covered.

    On screen Machine value Meaning
    Policy blocked (Hold) Hold A policy rule blocks this release until the blocking finding is fixed or a risk is accepted.
    Passed with accepted risk (Go with accepted risk) Go with accepted risk Nothing blocks the release because someone accepted the remaining risk; the acceptance is recorded.
    Checks passed (Go) Go The policy checks passed for what this analysis covered. It is not a statement that the code is safe.
    Pending Pending Analysis is still running; the decision follows when it completes.
    Not assessed Not assessed No policy decision was made for this target (it was not scanned, or no policy applies).

    Dense tables drop the bracketed machine value once the view has spelled it out in full. See release decisions and policies.

    How much the analysis covered. Coverage is not safety, so these states never use the colour of a passed decision.

    On screen Meaning
    Analysis complete Every stage that applies to this target ran to the end.
    Analysis incomplete Some files or stages were skipped; findings may be missing. Run details say what.
    Analysing The analysis is still running; counts can still change.
    Not assessed This was not analysed, so nothing is known about it either way.
    No baseline There is no earlier analysis to compare with, so nothing can be called new or fixed yet.

    The CLI reports an incomplete run with exit code 3. See coverage and limits.

    How bad a finding is if it is real. Severity says nothing about how sure Keryn is; that is evidence, below.

    On screen Meaning
    Critical Exploitable for serious harm if the finding is real; fix before release.
    High Likely harmful if real; fix soon.
    Medium Harmful in some conditions; plan a fix.
    Low Limited harm; fix when convenient.
    Info An observation, not a vulnerability.

    Rules state one of three engine levels, and SARIF and JSON carry those: ERROR is shown as High, WARNING as Medium and INFO as Info.

    How much Keryn established about a finding. Every level is drawn in the same colour: stronger evidence can support a more serious vulnerability, so it is never shown as “greener”. Strength is in the words and the filled marks.

    On screen Proof level Meaning
    Inventory 0 An observation of what is there (a model, a key, a dependency), not a vulnerability claim.
    Pattern match 1 The code matches a risky pattern. No data flow was traced.
    Within-file flow 2 Data was traced from a source to this place inside one file.
    Cross-file flow 3 Data was traced from a source to this place across files.
    Known-fix match 4 A published advisory or fix names this code. It corroborates the finding; it is not proof it is exploitable here.
    Reproduced 5 The behaviour was reproduced in a recorded environment.

    A known-fix match is not stronger than a traced flow, although its number is higher: it says an advisory names the code, not that data reaches it. The CLI’s text output leads with the name and keeps the number in brackets; SARIF and JSON carry the number. The proof ladder explains each level with examples.

    Word Meaning
    Initial checks The first pass over the code: the pattern rules. Engine name: the pattern engine.
    Cross-file analysis The pass that follows data across files and confirms or questions the pattern findings. Engine name: Deep; the flag is --no-deep.
    Edge One thing reaching another: user input into a prompt, model output into a shell, an agent into a tool.
    Trust boundary An edge where authority changes. See why the edges.
    Baseline An earlier analysis of the same target that a new one is compared with, to say what is new and what is fixed.
    Accepted risk A finding someone decided to release with. The acceptance names who, why and until when, and is recorded.