Sweep

Sweep

Three parallel auditors, each focused on one lens. Consolidated findings by severity. No character voice.

Announce every phase. "Dispatching 3 parallel auditors: security, stubs, code quality...", "All audits complete. Consolidating findings..."

Read .claude/references/language.md if it exists. Conduct all operator-facing conversation in the interface language. Auditor subagents operate in English. If the file doesn't exist, default to English.

Read .claude/references/verbosity.md if it exists. If level: minimal, announce dispatch and consolidated results only — skip per-auditor progress updates. If level: verbose or the file doesn't exist, narrate each auditor's progress. Default to minimal.

Before Starting

1. Determine scope. If the operator specified directories or file patterns, scope the audit to those. Otherwise, audit the entire project.
2. Skip non-code. Exclude node_modules/, vendor/, .git/, build artifacts, and binary files.

Dispatch

Dispatch all 3 auditors in a single message (parallel):

1. Security auditor — read references/security-auditor-prompt.md, fill {SCOPE} with the target files/directories, dispatch via Agent with subagent_type: "general-purpose".
2. Stubs auditor — read references/stubs-auditor-prompt.md, same dispatch pattern.
3. Code quality auditor — read references/quality-auditor-prompt.md, same dispatch pattern.

Each auditor returns a structured list of findings with file paths, line numbers, descriptions, and suggested fixes.

Consolidate

After all 3 auditors complete:

1. Deduplicate. Same file + same line + overlapping description = one finding. Keep the most specific version.

2. Classify by severity:

   • Critical — security vulnerabilities, exposed secrets, authentication bypass
   • High — stub implementations reachable in production, authorization gaps, empty error handlers in critical paths
   • Medium — code quality issues, minor inconsistencies, non-critical dead code
   • Low — style issues, minor cleanup, informational

3. Present to operator in severity order. Each finding includes:

   • Severity tag
   • Source auditor (security / stubs / quality)
   • File path and line number
   • Description
   • Suggested fix

Output Format

## Audit Results

### Critical (N)
1. **[security]** `src/auth/login.ts:45` — SQL injection in user lookup. Use parameterized query.

### High (N)
1. **[stubs]** `src/api/payments.ts:123` — Returns hardcoded `{status: "ok"}`. Stub reachable in production.

### Medium (N)
...

### Low (N)
...

**Summary:** N critical, N high, N medium, N low across N files.

If no findings in a severity category, omit that section.

Standalone vs Post-Execute

Standalone: Operator invokes /sweep directly. Run on current project state.

Post-execute: Offered by gigo:execute after all tasks complete. Can scope to only files changed during execution (via git diff from first task commit to HEAD).