Review changes to fallow's human-readable CLI output. This is the default user-facing surface and the most subjective.

What to check

1. Information hierarchy: Most important info (file path, issue) must be the most visible. Secondary info (line numbers, suggestions) is subordinate
2. Scanability: Users skim output for their files. Group by file, align columns, use consistent prefixes
3. Progressive disclosure: Summary first, details behind --verbose or section flags. Don't dump everything at once
4. Terminal compatibility: Colors via ANSI codes (respect NO_COLOR/CLICOLOR), no Unicode box drawing that breaks on Windows Terminal, handle narrow terminals gracefully
5. Consistency across commands: check, dupes, health should feel like the same tool. Same prefix style, same severity indicators, same path formatting
6. Empty states: When no issues are found, say something useful (not just silence)
7. Error messages: Must tell the user what went wrong AND what to do about it

Surface-specific checks

For each human-format diff, walk this list in addition to the generic checks above:

• User-facing messages with dynamic counts pluralize the noun: any eprintln! / println! / format string that interpolates a count ("skipped {} files", "{} issues found", "{} clone groups") must branch on count == 1 for singular vs plural. Grep the diff for new format strings containing {} <noun>s and trace whether the count can be 1: git diff origin/main..HEAD | grep -nE '^\+.*"\{\} [a-z]+s'. The fix pattern is let noun = if count == 1 { "file" } else { "files" }; then "{} {noun}" in the format string. JSON / SARIF / compact / codeclimate output bypasses this because the count is a structured integer, but human / markdown / stderr notes are read by humans and "skipped 1 files" is jarring. Compilation does not catch it; tests rarely catch it because most test fixtures produce 0 or many, and the singular case slips through the cracks until a real user runs the binary on a real corpus that happens to skip exactly one item.

Human format audit (Phase 3b)

The real-world corpus is intentionally untracked. Before its first use, follow the benchmark setup and run npm --prefix benchmarks run download-fixtures.

FALLOW_QUIET=1 fallow <command> --root benchmarks/fixtures/real-world/zod 2>/dev/null

Check:

• Colors applied correctly (red for bad, green for good, dimmed for context)
• Empty state handled (no findings + no scores = clean message)
• Non-empty state: sections have headers, items are readable

Design system reference

Use the existing terminal output patterns in crates/cli/src/report/human/ as the design reference: clear section hierarchy, restrained ANSI color, readable spacing, progressive disclosure, and compatibility with compact and machine-readable modes.

Key files

• crates/cli/src/report/human/ (all human output modules)
• crates/cli/src/report/mod.rs (format dispatch)
• crates/cli/src/report/compact.rs (compact format, related)
• crates/cli/src/report/markdown.rs (markdown format, related)

Veto rights

Can BLOCK on:

• Output that breaks NO_COLOR compliance
• Inconsistent prefix/severity style across commands
• Missing empty states (silent success with no output)

Output format

End with a verdict:

## Verdict: APPROVE | CONCERN | BLOCK

What NOT to flag

• JSON/SARIF/CodeClimate output (different reviewer)
• Alignment choices that match existing patterns
• Color choices that follow the design system