Reading the report

This shortened excerpt is from a real Claude Code 0.2.0 live run on 22 October 2026. It omits local paths, IDs, timings and the full task text; the lines shown retain the report’s wording.

Verdict: PASS
Status: PASS (verify PASS; steered 1 time, 1 blocked)

## What changed
- src/greet.js (+1, −1)
- src/shout.js (+1, −1)

## Tests
- npm test: passed (exit 0; verify, isolated copy); 2 passed, 0 failed, 0 skipped.

## Runtime check
- Passed — node 'bin.js' --help: CLI runtime probe completed.

Before that PASS, the Stop gate ran npm test, saw a failure and asked the agent to continue. The contract later allowed src/shout.js because a failing protected test imported it. This is one observed run, not a reliability rate.

What the labels mean

DONE

A requested check ran and passed.

EXTRA

A changed path went beyond the allowed scope; read whether it blocks the verdict.

TAMPERING

Tests or protected state may have been weakened or altered.

OBSERVED

A changed file was seen by hooks and checked by the test command, without a scenario declaring coverage.

INCOMPLETE

The overall verdict cannot establish a pass or fail because required evidence is missing. Read the specific gap in the report.

RUNTIME

A user-like CLI, server or browser probe, compared with the base when it ran.

SUPERVISOR

Recorded block, warn and ask decisions; hook output alone does not prove execution.

The full report also has CHECKED, NOT RUN, POLICY, capture coverage, metrics and evidence tags. “Verified in this environment” means a command ran there; “Observed” means it came from a local record; “Unknown” means no result was established. PASS is bounded by those distinctions.