tune-multi-agent-service — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited tune-multi-agent-service (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.
| Parameter | Default | Range | Impact |
|---|---|---|---|
| Routing model | gpt-4o | gpt-4o / gpt-4o-mini | Quality vs speed of routing |
| Max subtasks | 5 | 2-10 | More = thorough, fewer = faster |
| Parallel execution | Disabled | On/Off | Parallel = faster, sequential = more context |
| Routing temperature | 0.0 | 0.0-0.1 | Must be deterministic |
| Routing few-shot examples | 3 | 2-5 | More = better routing accuracy |
Diagnostic: Check routing accuracy with python evaluation/eval.py --metrics routing
| Agent Role | Default | Cost-Optimized | Quality-Optimized |
|---|---|---|---|
| Supervisor | gpt-4o | gpt-4o (keep) | gpt-4o (keep) |
| Researcher | gpt-4o-mini | gpt-4o-mini | gpt-4o |
| Analyst | gpt-4o | gpt-4o-mini | gpt-4o |
| Writer | gpt-4o-mini | gpt-4o-mini | gpt-4o |
| Validator | gpt-4o-mini | gpt-4o-mini | gpt-4o-mini |
Rule: Supervisor always uses the most capable model. Workers can be downgraded for cost.
| Agent | Default | Min | Max | Diagnostic |
|---|---|---|---|---|
| Supervisor | 30s | 15s | 60s | Task decomposition time |
| Researcher | 20s | 10s | 40s | Search + retrieval time |
| Analyst | 25s | 15s | 45s | Analysis complexity |
| Writer | 15s | 8s | 30s | Content generation length |
| Validator | 10s | 5s | 20s | Validation check count |
Optimization: Set timeout = p95 observed latency + 20% buffer
| Strategy | Context Size | Use Case |
|---|---|---|
| Full context | All prior agent outputs | Complex reasoning chains |
| Summary only | Supervisor summarizes before handoff | Long chains, token savings |
| Key-value store | Only relevant state passed | High-volume, cost-sensitive |
| Streaming | Token-by-token from prior agent | Real-time applications |
Cost impact: Full context can use 4x more tokens than summary-only mode.
Baseline cost (5-agent chain, all gpt-4o, full context):
Optimized (mixed models, summary context):
| Parameter | Default | Stricter | Looser |
|---|---|---|---|
| Max handoffs per task | 10 | 5 | 15 |
| Max cost per task | $0.50 | $0.10 | $1.00 |
| Circuit breaker threshold | 3 failures | 2 failures | 5 failures |
| Human escalation trigger | 2 retries | 1 retry | 3 retries |
After tuning, generate comparison:
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.