VibeGuard — first win

FIRST WIN / HONEST READ

Next action{{NEXT_ACTION}}

Evidence in two tiers

Headline evidence comes only from observe value --json.

VERIFIED ASSOCIATION

Time order in one non-empty session; not proof of cause.

Later build pass after intervention

OBSERVED SIGNALS

What the selected local event stream contains.

Follow-up after intervention

Sessions without later follow-up

Observed hook overhead

OBSERVED FRICTION / NOT OUTCOME EVIDENCE

{{REPEATED}}sessions with repeated guardrail signals

{{SUPPRESSIONS}}suppressed guardrail signals

{{UNCORRELATABLE}}signals missing session/time correlation

ESTIMATED / UNAVAILABLE

BOUNDARY / BUILT-IN CHECK

The built-in 5-positive/5-negative benchmark is a small reproducible corpus, not evidence of impact on your project.

Keep benchmark behavior checks separate from this local event-stream evidence.

Secondary context for troubleshooting. Raw local output is shown only inside expandable details and is escaped before rendering.

Codex installation status

{{STATUS_OUT}}

{{STATUS_COMMAND}}

Hook health output

{{HEALTH_OUT}}

{{HEALTH_COMMAND}}

Trigger stats output

{{STATS_OUT}}

{{STATS_COMMAND}}

Observe-value command output

{{VALUE_OUT}}

{{VALUE_COMMAND}}