Skip to content

Exit codes & contracts ​

Frozen — safe to build CI logic on:

Exit codeMeaning
0Clean — complete analysis, no findings at or above the gate
1Findings at or above the gate
2Inconclusive — the analysis did not finish, or the surface is unsupported. Treat as failure.
10Usage error (bad flag, missing target)
20Internal error

These are scan/CI exits. ci release-trust and ci release-trend use their own verdicts and exit contracts.

Exit 2 is not a pass ​

Earlier versions of this page said exit 2 "never blocks". That was wrong, and it was wrong in the direction that ships: a pipeline that treated an incomplete analysis as success reported a clean result for a surface nobody had finished analyzing.

2 means one of two things, and neither is a pass:

  • Partial — the time budget hit, files were unreadable or skipped, rules crashed, or discovery was truncated. The unanalyzed surface is exactly where an unknown blocking finding would live.
  • Unsupported or unmeasured — the command has no measurement of what you asked it about, and says so instead of printing a plausible number.

A CI step should fail on 2 the same way it fails on 1. The default for any non-zero exit already does; the point of stating it is that people had been told to special-case 2 out.

--blocking none (or gate: "advisory") suppresses findings, not the fact that the analysis did not finish. No flag turns an incomplete run into exit 0.

exit 0exit 2
Analysis covered the whole surfaceyesno
Proves the surface is cleanyesno

The single table that decides this lives in src/claim-evidence.ts (EXIT_MATRIX); scanExitCode is the only function that consults it.

The JSON/SARIF report is schemaVersion: 1. Rule IDs (QA-<FAMILY>-NNN) are immutable once shipped and never reused.

Trust model ​

  • Local-first — zero network calls during scanning. Ever. Zero telemetry.
  • No false proof — an empty repo gets score: null, never a fake 100.
  • Partial honesty — if analysis was cut short, the output says so, and the exit code is 2. Honesty that still exits 0 is not honesty.
  • FP firewall — detection runs on a comment/string-free view of the code.
  • Measured, not asserted — 73 of 79 active rules carry a real OSS false-positive rate; the other 6 are explicitly unmeasured.
  • Plugin trust — plugins are npm packages with no sandbox; they run with full Node privileges, the same trust model as ESLint or Vitest plugins.

Released under the MIT License.