tune-content-moderation — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited tune-content-moderation (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
| Category | Consumer App | Enterprise Internal | Developer Tool |
|---|---|---|---|
| Hate | ≥ 2 (Low) | ≥ 4 (Medium) | ≥ 4 (Medium) |
| Violence | ≥ 2 (Low) | ≥ 4 (Medium) | ≥ 6 (High) |
| SelfHarm | ≥ 2 (Low) | ≥ 2 (Low) | ≥ 2 (Low) |
| Sexual | ≥ 2 (Low) | ≥ 4 (Medium) | ≥ 4 (Medium) |
Tuning methodology:
| Strategy | Impact | Complexity |
|---|---|---|
| Raise severity threshold | Reduces FP, may miss some harmful | Low |
| Add context exceptions | Domain-specific exclusions | Medium |
| Custom categories | Replace broad detection with targeted | High |
| Human-in-the-loop | Review borderline cases | Operational |
Context-aware exceptions:
| Parameter | Default | Recommendation | Notes |
|---|---|---|---|
| Term count per blocklist | 0 | 50-500 | Too many = performance overhead |
| Blocklist count | 0 | 2-5 per domain | Separate by category |
| Refresh frequency | Manual | Weekly | Stakeholder review before adding |
| Partial matching | Disabled | Keep disabled | Reduces false positives |
Blocklist hygiene:
| Strategy | Latency Reduction | When to Use |
|---|---|---|
| Text chunking | 30-50% for long text | Documents > 1000 chars |
| Parallel analysis | 40-60% for batch | Multiple content items |
| Category subset | 20-30% | Only check relevant categories |
| Caching | 80-95% for repeated | FAQ/template content |
| Regional deployment | 10-30ms | Cross-region latency |
Optimization steps:
| Component | Cost Driver | Optimization |
|---|---|---|
| Text analysis | 1,000 text records per unit | Batch requests, use caching |
| Image analysis | 1 image per unit | Resize before analysis, skip thumbnails |
| Custom categories | Training + inference | Use only when prebuilt insufficient |
| Blocklists | Free (included) | Prefer blocklists over custom categories |
Monthly cost estimate (100K moderations/day):
| Profile | Thresholds | Blocklists | Latency Target |
|---|---|---|---|
| Consumer chatbot | Strict (≥ 2) | Brand + competitor | < 100ms |
| Enterprise copilot | Moderate (≥ 4) | Company terms | < 200ms |
| Content generation | Strict (≥ 2) | All active | < 150ms |
| Code assistant | Relaxed (≥ 6) | None | < 50ms |
After tuning, compare before/after:
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.