/deslop - AI Slop Cleanup

Senior maintainer performing periodic repo hygiene. Mission: remove AI-generated slop while preserving behavior and minimizing diffs.

Constraints (Priority Order)

When constraints conflict, follow this priority:

1. Preserve behavior and public APIs (highest priority)
2. Minimal diffs - do not reformat unrelated code
3. Prefer deletion over invention
4. No new dependencies or abstractions
5. Respect repo conventions (check CLAUDE.md/AGENTS.md)

Arguments

Parse from $ARGUMENTS or use defaults:

• Mode: report (default) or apply
• Scope: all (default), diff, or path
• Thoroughness: quick, normal (default), or deep

Execution

Phase 1: Spawn Deslop Agent

// Parse arguments
const args = '$ARGUMENTS'.split(' ').filter(Boolean);
const mode = args.includes('apply') ? 'apply' : 'report';
const thoroughness = args.find(a => a.startsWith('--thoroughness='))?.split('=')[1] || 'normal';
const scope = args.find(a => a.startsWith('--scope='))?.split('=')[1] ||
              args.find(a => !a.startsWith('-') && a !== 'report' && a !== 'apply') || 'all';

// Spawn agent to get findings
const result = await Task({
  subagent_type: "deslop:deslop-agent",
  prompt: `Scan for AI slop patterns.
Mode: ${mode}
Scope: ${scope}
Thoroughness: ${thoroughness}

Return structured results between === DESLOP_RESULT === markers.`
});

Phase 2: Parse Agent Results

Extract structured JSON from agent output:

function parseDeslop(output) {
  const match = output.match(/=== DESLOP_RESULT ===[\s\S]*?({[\s\S]*?})[\s\S]*?=== END_RESULT ===/);
  return match ? JSON.parse(match[1]) : { fixes: [] };
}

const findings = parseDeslop(result);

Phase 3: Handle Mode

Report Mode (Default)

Present findings as markdown table:

## Slop Hotspots

| Priority | File | Issue | Certainty | Fix |
|----------|------|-------|-----------|-----|
| 1 | src/api.js:42 | console.log | HIGH | auto |
| 2 | src/auth.js:15 | empty catch | HIGH | auto |
| 3 | lib/utils.js:88 | excessive comments | MEDIUM | review |

## Summary

- **HIGH certainty**: N (auto-fixable)
- **MEDIUM certainty**: N (review required)
- **LOW certainty**: N (flagged only)

## Do Next

- [ ] Run `/deslop apply` to auto-fix HIGH certainty items
- [ ] Review MEDIUM certainty items manually

Apply Mode

If fixes array is non-empty, spawn simple-fixer:

if (mode === 'apply' && findings?.fixes?.length > 0) {
  await Task({
    subagent_type: "next-task:simple-fixer",
    model: "haiku",
    prompt: `Apply these slop fixes:
${JSON.stringify(findings.fixes, null, 2)}

For each fix:
- remove-line: Delete the line at the specified line number
- add-comment: Add "// Error intentionally ignored" to empty catch
- remove-block: Delete the entire code block

Use Edit tool to apply. Commit message: "fix: clean up AI slop (auto-applied)"`
  });
}

Present results:

## Applied Fixes

| File | Line | Fix |
|------|------|-----|
| src/api.js | 42 | remove-line (console.log) |
| src/auth.js | 15 | add-comment (empty catch) |

**Total**: N fixes applied

## Remaining (manual review needed)

| File | Line | Issue | Certainty |
|------|------|-------|-----------|
| lib/utils.js | 88 | excessive comments | MEDIUM |

Verification Strategy

After fixes are applied, run project's test command:

npm test
# or pytest, cargo test, go test ./...

On failure: git restore . and report which change failed.

Ignore Zones

Skip these paths (handled by detection script):

• Build artifacts: dist/, build/, target/, out/, .next/
• Vendored: vendor/, node_modules/, **/*.min.*
• Generated: **/*.gen.*, lockfiles

Error Handling

• Git not available: Exit with "Git required for rollback safety"
• Invalid scope path: Exit with "Path not found: <path>"
• Verification fails: Rollback with git restore ., report failure

Additional Resources

Reference Files

For detailed pattern documentation, consult:

• references/slop-categories.md - All pattern categories, severity levels, certainty thresholds, auto-fix strategies

Scripts

• scripts/detect.js - Detection pipeline CLI (run with --help for options)