The repo needed a first-class CLI surface for scoring session health without relying on side scripts. This threads a native report subcommand through the existing help, parsing, and shared context paths and locks the behavior with focused regression coverage.

Constraint: Keep the scoring flow inside the existing report command surface and T1 notes-only OpenSpec lane
Rejected: Ship a standalone side script | would drift from CLI parsing and help text
Confidence: high
Scope-risk: narrow
Directive: Keep help text, rubric weights, and report tests synchronized when the scoring model changes
Tested: node --test test/cli-args-dispatch.test.js test/report.test.js; node bin/multiagent-safety.js report help; node bin/multiagent-safety.js report session-severity --task-size narrow-patch --tokens 3850000 --exec-count 18 --write-stdin-count 6 --completion-before-tail yes --fragmentation 14 --finish-path 6 --post-proof 4; git diff --check
Not-tested: remote PR merge path until guarded finish completes