nwave-ai/nw-hotspot

NW-HOTSPOT: Code Crime Scene Hotspot Lens

Wave: CROSS_WAVE Execution: Inline (no agent — Claude executes git commands directly) Inspiration: Adam Tornhill's "Your Code as a Crime Scene"

Overview

Analyze git change frequency to identify the most-changed files in a codebase. Pure git churn — no complexity metrics. Use as a pre-filter to scope analysis or a post-filter to prioritize existing findings.

How It Works

Run git log --name-only over the configured time period, count commits per file, rank by frequency.

Modes

1. Analyze (default)

Show the most-changed files in the repo. Output is inline only — no files persisted.

/nw-hotspot
/nw-hotspot --top=20
/nw-hotspot --since=12m

Steps:

1. Run: git log --since={since} --format=format: --name-only | sort | uniq -c | sort -rn
2. Exclude deleted files (verify each file still exists with test -f)
3. For each file, get last changed date: git log -1 --format="%ar" -- {file}
4. Display ranked table (top N):

 Rank │ Commits │ Last Changed │ File
──────┼─────────┼──────────────┼─────────────────────────
    1 │      87 │ 2 days ago   │ src/services/payment.ts
    2 │      64 │ 1 week ago   │ src/api/controllers/user.ts
    3 │      51 │ 3 days ago   │ src/models/order.ts

5. Show summary: total files changed in period, time period, top N shown

Defaults: --top: 15, --since: 6 months

2. Rank (post-filter)

Overlay churn data on an existing analysis report to prioritize findings.

/nw-hotspot --rank code-smell-detector-report.md

Steps:

1. Run the same git churn analysis
2. Read the specified report file
3. Extract file paths mentioned in the report
4. Annotate each file with its commit count
5. Re-sort findings by churn (highest first). Findings without extractable file paths go at the bottom.
6. Show noise reduction: "X of Y findings are in low-churn files (< 20 commits)"

3. Detail (deep-dive)

Deep-dive into a single file's change history.

/nw-hotspot --detail src/services/payment.ts

Steps:

1. Monthly commit breakdown over the period
2. Top contributors to the file
3. Co-change coupling: top 5 files that frequently change alongside this one
4. Last 10 commit messages for context

4. JSON export (pre-filter composition)

/nw-hotspot --top=10 --json

Output: [{"path": "src/services/payment.ts", "commits": 87, "last_changed": "2 days ago"}, ...]

Parameters

| Flag | Default | Description | |------|---------|-------------| | --top | 15 | Number of files to show | | --since | 6m | Time period (e.g., 3m, 12m, 2024-01-01) | | --json | false | Output as JSON array | | --rank | - | Path to existing report to re-rank by churn | | --detail | - | Path to single file for deep-dive |

Edge Cases

• No commits in period: Display "No commits found since {date}" rather than an empty table
• File renames: Counted as separate files (no --follow; keeps it simple and fast)
• Empty repo / not a git repo: Display error and exit

Usage from Other Skills

Pre-filter (scope downstream analysis)

Run /nw-hotspot --top=10 --json first, then pass the file list to:

• code-smell-detector: "Analyze only these files: {hotspot list}"
• cognitive-load-analyzer: "Focus analysis on these files: {hotspot list}"
• refactoring-expert: "Prioritize recommendations for these files"

Post-filter (rank existing findings)

Run /nw-hotspot --rank {report.md} after any analysis to prioritize by churn.

In nWave workflows

• Before /nw-refactor — identify which files to refactor first
• Before /nw-review — focus review effort on high-churn areas
• Before /nw-root-why — check if the problematic area is a known hotspot

Next Wave

Usage Context: Inline utility, composable with any analysis skill Handoff To: Invoking context (no sequential handoff)

Examples

Example 1: Quick scan of a legacy codebase

/nw-hotspot

Example 2: Scope a code smell analysis to hotspots

/nw-hotspot --top=10 --json

Example 3: Prioritize an existing refactoring report

/nw-hotspot --rank code-refactoring-report.md

Example 4: Investigate a specific troubled file

/nw-hotspot --detail src/services/payment.ts

Expected Outputs

No persistent files. Output is displayed inline. JSON mode outputs a JSON array for composition with other tools.