Is tune-cost-optimized-ai-gateway safe to install?

SaferSkills independently audited tune-cost-optimized-ai-gateway (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.

CategoryWeightCategory scoreContribution

Securityprompt, exec, net, exfil, eval

35%

100

35.0 pts

Supply chainhash, typosquat, maintainer, lockfile

20%

100

20.0 pts

Maintenancestaleness, pinning, CI

15%

100

15.0 pts

TransparencySKILL.md, perms, README

15%

100

15.0 pts

Communityinstalls, verify, response

15%

100

15.0 pts

Findings & checks · 0 flagged

Securityscore 100 · 0 findings

✓—All security checks passedNo findings in this category for the latest scan.pass

Supply chainscore 100 · 0 findings

✓—All supply chain checks passedNo findings in this category for the latest scan.pass

Maintenancescore 100 · 0 findings

✓—All maintenance checks passedNo findings in this category for the latest scan.pass

Transparencyscore 100 · 0 findings

✓—All transparency checks passedNo findings in this category for the latest scan.pass

Communityscore 100 · 0 findings

✓—All community checks passedNo findings in this category for the latest scan.pass

The primary manifest — the file an agent reads to learn what this artifact does.

solution-plays/14-cost-optimized-ai-gateway/.github/skills/tune-cost-optimized-ai-gateway/SKILL.md · 1 file

solution-plays/14-cost-optimized-ai-gateway/.github/skills/tune-cost-optimized-ai-gateway/SKILL.md4.7 KB · Markdown

name: tune-cost-optimized-ai-gateway

description: "Tune AI Gateway — optimize cache TTL and similarity threshold, PTU vs pay-as-you-go, routing weights, budget tiers, cost per 1K tokens. Use when: tune, optimize, FinOps."

Tune Cost-Optimized AI Gateway

When to Use

Optimize semantic cache parameters for maximum savings
Decide PTU (Provisioned Throughput Units) vs pay-as-you-go
Tune multi-region routing weights based on traffic patterns
Configure budget tiers and rate limits per tenant
Minimize cost per 1K tokens through the gateway

Tuning Dimensions

Dimension 1: Semantic Cache Optimization

Parameter	Default	Range	Impact
Similarity threshold	0.95	0.85-0.99	Lower = more hits, risk of wrong cached answer
Cache TTL	1 hour	5min-24hr	Longer = more hits, risk of stale data
Max cache entries	100K	10K-1M	More = more VRAM, higher hit rate
Embedding model	text-embedding-3-small	small/large	Small = cheaper + faster cache lookup
Cache scope	Per-tenant	Global/Per-tenant	Global = higher hit rate, shared privacy risk

Tuning methodology:

Start with threshold=0.95, TTL=1h
Monitor cache hit rate for 7 days
If hits <20%: lower threshold to 0.92, increase TTL
If stale responses reported: reduce TTL, raise threshold
Target: 30-50% hit rate = 30-50% cost savings

Dimension 2: PTU vs Pay-As-You-Go Decision

Factor	PTU	Pay-As-You-Go
Pricing	Fixed monthly	Per-token
Best for	Predictable, high-volume	Variable, bursty
Latency	Guaranteed	Best-effort
Commitment	1-month minimum	None
Break-even	~30K TPM sustained	Below 30K TPM

Decision matrix:

Monthly Volume	Recommendation	Why
<1M tokens/day	Pay-as-you-go	PTU overhead not justified
1M-10M tokens/day	PTU for base + PAYG for burst	Hybrid = cost + flexibility
>10M tokens/day	PTU (committed)	Significant savings + guaranteed throughput

Dimension 3: Routing Weight Optimization

Strategy	Configuration	When to Use
Weighted round-robin	60/40 primary/secondary	General traffic distribution
Latency-based	Route to fastest region	Latency-sensitive applications
Cost-based	Route to cheapest region	Pure FinOps optimization
Capacity-based	Route to lowest utilization	Prevent throttling
Compliance-based	Route by data residency	GDPR/HIPAA requirements

Dynamic weight tuning:

Monitor TPM utilization per region
If region A >80% TPM: shift 20% weight to region B
If region B has lower pricing tier: shift non-latency-critical traffic there

Dimension 4: Budget Tier Configuration

Tier	Budget	Rate	Model	% of Users
Free	100K tok/mo	10/min	mini only	60%
Standard	1M tok/mo	60/min	mini + 4o	30%
Premium	10M tok/mo	300/min	All	8%
Enterprise	Unlimited	1000/min	All + priority	2%

Cost allocation: Track per-tenant usage in Redis sorted sets for O(1) budget checks.

Dimension 5: End-to-End Cost Optimization

Monthly cost breakdown (100K requests/day, avg 500 tokens/request):

Component	Without Gateway	With Gateway (30% cache)	Savings
Azure OpenAI (gpt-4o)	$7,500	$5,250	$2,250
APIM (Standard v2)	—	$300	—
Redis (Premium P1)	—	$400	—
Embedding (cache key)	—	$20	—
Total	$7,500	$5,970	$1,530 (20%)

With 50% cache hit: Total drops to $4,470 — $3,030 savings (40%).

Optimization levers:

Increase cache hit rate (threshold + TTL tuning)
Route simple queries to gpt-4o-mini (80% cheaper)
Use PTU for base load (30% cheaper than PAYG)
Enforce budgets to prevent runaway costs
Compress prompts before caching (reduces Redis memory)

Production Readiness Checklist

[ ] Cache hit rate ≥ 30% observed over 7 days
[ ] Cost savings ≥ 20% vs direct API calls
[ ] Gateway overhead < 50ms p95
[ ] PTU vs PAYG decision documented with break-even analysis
[ ] Budget tiers enforce correctly at boundaries
[ ] Multi-region failover tested (< 3s switchover)
[ ] Per-tenant cost tracking accurate (< 5% error)
[ ] Rate limiting active and tested
[ ] FinOps dashboard showing real-time cost allocation
[ ] Budget alerts at 80%/90%/100% thresholds

Output: Tuning Report

After tuning, compare:

Cache hit rate improvement
Cost per 1K tokens reduction (before vs after)
Gateway overhead change
PTU recommendation with break-even analysis
Budget tier adjustment recommendations

tune-cost-optimized-ai-gateway— agent skill

Is tune-cost-optimized-ai-gateway safe to install?

Tune Cost-Optimized AI Gateway

When to Use

Tuning Dimensions

Dimension 1: Semantic Cache Optimization

Dimension 2: PTU vs Pay-As-You-Go Decision

Dimension 3: Routing Weight Optimization

Dimension 4: Budget Tier Configuration

Dimension 5: End-to-End Cost Optimization

Production Readiness Checklist

Output: Tuning Report

Audit the pieces. Scan the whole. Decide.