Is tune-agentic-rag safe to install?

SaferSkills independently audited tune-agentic-rag (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/21-agentic-rag/.github/skills/tune-agentic-rag/SKILL.md · 1 file

solution-plays/21-agentic-rag/.github/skills/tune-agentic-rag/SKILL.md4.1 KB · Markdown

name: tune-agentic-rag

description: "Tune Agentic RAG — optimize source routing weights, iteration thresholds, self-evaluation sensitivity, semantic cache, cost per query. Use when: tune, optimize retrieval."

Tune Agentic RAG

When to Use

Optimize source routing (which source for which query type)
Tune iteration limits and self-evaluation thresholds
Configure semantic cache for maximum hit rate
Reduce cost per query while maintaining quality
Balance retrieval depth vs response latency

Tuning Dimensions

Dimension 1: Source Routing Optimization

Source	Default Priority	When to Use	Cost per Call
AI Search (KB)	1 (primary)	Internal knowledge, policy, docs	~$0.003
Bing Web Search	2 (fallback)	Current events, external info	~$0.005
SQL Database	3 (specific)	Customer data, order status	~$0.001
Custom API	4 (specialized)	Domain-specific real-time data	Varies

Routing optimization:

Add query classifiers to route directly (skip unnecessary sources)
Cache source selection decisions for similar query patterns
Monitor: which sources are called but results never used? → deprioritize

Dimension 2: Iteration Threshold Tuning

Parameter	Conservative	Balanced	Aggressive
`max_retrieval_hops`	5	3	1
`min_groundedness`	0.95	0.85	0.75
`early_stop_confidence`	0.98	0.92	0.85
Quality	Highest	Good	Fastest
Cost	Highest	Moderate	Lowest
Latency	Highest	Moderate	Lowest

Tuning methodology:

Start balanced (3 hops, 0.85 groundedness)
If quality too low: increase hops and raise groundedness threshold
If cost too high: reduce hops, accept slightly lower quality
Track: queries that exhaust max hops (indicate poor source coverage)

Dimension 3: Self-Evaluation Tuning

Setting	Description	Impact
Strict (0.95)	Agent almost always iterates	High quality, high cost
Moderate (0.85)	Agent stops when reasonably grounded	Good balance
Relaxed (0.75)	Agent stops early, faster responses	Lower quality, lower cost

Anti-patterns:

Self-eval too strict → agent always uses max hops (expensive)
Self-eval too relaxed → agent responds with insufficient context (low quality)
Solution: calibrate on test set where you know the correct number of hops

Dimension 4: Semantic Cache Optimization

Parameter	Default	Range	Impact
Similarity threshold	0.92	0.85-0.98	Lower = more hits, risk wrong cache
TTL	1 hour	5min-24hr	Longer = more hits, stale risk
Max entries	50K	10K-500K	More = higher Redis cost
Embedding model	text-embedding-3-small	small/large	Small = faster lookup

Expected savings: 40-60% cost reduction on repeated query patterns.

Dimension 5: Cost Per Query Optimization

Scenario	Hops	Sources	Cache	Cost
Simple FAQ (cache hit)	0	0	✅	~$0.00
Simple FAQ (cache miss)	1	KB only	No	~$0.01
Standard query	2	KB + 1 more	No	~$0.03
Complex multi-source	3	KB + Web + DB	No	~$0.05
Max iteration	5+	All sources	No	~$0.10+

Monthly estimate (10K queries/day):

Without optimization: ~$15,000/mo (all queries 3 hops)
With cache (40% hit): ~$9,000/mo (40% savings)
With routing + cache: ~$6,000/mo (60% savings)

Production Readiness Checklist

[ ] Source routing accuracy ≥ 90%
[ ] Average hops < 2.0 per query
[ ] Self-evaluation calibrated (matches actual quality)
[ ] Cache hit rate ≥ 40%
[ ] Cost per query < $0.05 average
[ ] Groundedness ≥ 0.90
[ ] Citations accurate (≥ 95%)
[ ] Max iteration limit enforced
[ ] Response latency < 10s at p95

Output: Tuning Report

After tuning, compare:

Source routing accuracy improvement
Average hops reduction
Cache hit rate and cost savings
Cost per query reduction
Quality delta (groundedness/relevance)

tune-agentic-rag— agent skill

Is tune-agentic-rag safe to install?

Tune Agentic RAG

When to Use

Tuning Dimensions

Dimension 1: Source Routing Optimization

Dimension 2: Iteration Threshold Tuning

Dimension 3: Self-Evaluation Tuning

Dimension 4: Semantic Cache Optimization

Dimension 5: Cost Per Query Optimization

Production Readiness Checklist

Output: Tuning Report

Audit the pieces. Scan the whole. Decide.