Review Code Complexity

Use cyclomatic and cognitive complexity findings as cues for architectural assessment. Completion means an evidence-backed disposition for every reviewed symbol, not a lower score or a warning-free command.

Follow the user's tool choices when provided; otherwise use the standard tools available in the environment. Read and apply the codebase-design skill for assessment and the program-design skill when proposing substantial changes. Resolve these skills through the environment's available skills; report an unavailable required skill as a prerequisite gap instead of claiming to have applied it.

1. Establish the Review Batch

1. Read the repository instructions and inspect the current worktree. Preserve existing changes and distinguish them from work authorized by this review.
2. Discover the configured cyclomatic and cognitive complexity diagnostics, their scope, thresholds, exclusions, suppression conventions, and command behavior. Use the repository's existing configuration rather than introducing thresholds or installing tooling.
3. Honor explicitly requested files, findings, or package scope. With no narrower request, obtain repository-wide findings, sort flagged files by repository-relative path, and select the first file awaiting review. Review every flagged symbol in that file; there is no per-symbol cap.
4. Treat multiple metric findings for the same symbol as one assessment while retaining each diagnostic record. Identify anonymous callbacks and nested functions by their enclosing symbol and location.
5. For explicitly named files, also inspect existing complexity suppressions and their underlying code. A clean diagnostic run does not replace that assessment. Bare invocation targets currently emitted findings rather than reopening existing suppressions.
6. Maintain the reviewed-file dispositions and remaining queue in the conversation. On a subsequent bare invocation, refresh the findings and advance past files already reviewed against unchanged code and configuration. Reassess affected dispositions when their supporting evidence changes. A new conversation has no assumed review history.

If metrics are not configured, report the missing prerequisite and stop automatic selection. An explicitly named file can still receive a qualitative architectural assessment; label measured complexity as unavailable. Installing or configuring metrics requires a separate request. If configured diagnostics fail to run, record the exact failure and incomplete coverage; use available source evidence for an explicitly scoped review without claiming a measured baseline.

Complete this phase when the selected files, diagnostic coverage, existing-change ownership, and remaining queue are recorded, or the exact selection blocker is established. If a complete run yields no findings, report that result without claiming the repository has no architectural problems.