Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Execute use when provisioning Vertex AI ADK infrastructure with Terraform. Trigger with phrases like "deploy ADK terraform", "agent engine infrastructure", "provision ADK agent", "vertex AI agent terraform", or "code execution sandbox terraform". Provisions Agent Engine runtime, 14-day code execution sandbox, Memory Bank, VPC Service Controls, IAM roles, and secure multi-agent infrastructure.
.claude/skills/jeremylongshore-adk-infra-expert/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-10 | ✗→✓ | ▲ Improved | -5% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 24% | 0% |
| case-09 | ✓→✗ | ▼ Worse | 13% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | 9% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 14% | 0% |
Provision production-grade Vertex AI ADK infrastructure with Terraform: secure networking, least-privilege IAM, Agent Engine runtime, Code Execution sandbox defaults, and Memory Bank configuration. Use this skill to generate/validate Terraform modules and a deployment checklist that matches enterprise security constraints (including VPC Service Controls when required).
Before using this skill, ensure:
See Terraform implementation details for output format specifications.
See ${CLAUDE_SKILL_DIR}/references/errors.md for comprehensive error handling.
See ${CLAUDE_SKILL_DIR}/references/examples.md for detailed examples.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 31,912 | 27,656 | -13% | 1 | 1 | 0% | 6,220 | 6,553 | +5% | 0 | 0 | — |
case-02 | fail→fail | 27,862 | 27,426 | -2% | 1 | 1 | 0% | 6,211 | 6,659 | +7% | 0 | 0 | — |
case-03 | fail→fail | 27,991 | 24,411 | -13% | 1 | 1 | 0% | 6,211 | 6,057 | -2% | 0 | 0 | — |
case-04 | pass→pass | 22,053 | 21,807 | -1% | 1 | 1 | 0% | 5,132 | 5,612 | +9% | 0 | 0 | — |
case-05 | pass→pass | 12,408 | 11,545 | -7% | 1 | 1 | 0% | 2,545 | 2,907 | +14% | 0 | 0 | — |
case-06 | pass→pass | 16,303 | 13,948 | -14% | 1 | 1 | 0% | 3,743 | 3,649 | -3% | 0 | 0 | — |
case-07 | pass→pass | 16,597 | 13,336 | -20% | 1 | 1 | 0% | 3,522 | 3,255 | -8% | 0 | 0 | — |
case-08 | pass→pass | 16,240 | 8,520 | -48% | 1 | 1 | 0% | 2,875 | 2,112 | -27% | 0 | 0 | — |
case-09 | pass→fail | 16,511 | 15,461 | -6% | 1 | 1 | 0% | 3,012 | 3,411 | +13% | 0 | 0 | — |
case-10 | fail→pass | 11,080 | 7,781 | -30% | 1 | 1 | 0% | 2,035 | 1,935 | -5% | 0 | 0 | — |
case-11 | pass→pass | 8,622 | 7,039 | -18% | 1 | 1 | 0% | 1,660 | 1,882 | +13% | 0 | 0 | — |
case-12 | fail→pass | 14,062 | 13,069 | -7% | 1 | 1 | 0% | 2,635 | 3,271 | +24% | 0 | 0 | — |
case-13 | pass→pass | 12,128 | 10,363 | -15% | 1 | 1 | 0% | 2,148 | 2,389 | +11% | 0 | 0 | — |
case-14 | pass→pass | 8,688 | 10,603 | +22% | 1 | 1 | 0% | 1,603 | 2,366 | +48% | 0 | 0 | — |
case-15 | pass→pass | 20,036 | 19,512 | -3% | 1 | 1 | 0% | 3,963 | 4,767 | +20% | 0 | 0 | — |
case-16 | pass→pass | 9,805 | 5,897 | -40% | 1 | 1 | 0% | 1,820 | 1,566 | -14% | 0 | 0 | — |
case-17 | pass→pass | 13,895 | 12,689 | -9% | 1 | 1 | 0% | 2,748 | 3,064 | +11% | 0 | 0 | — |
case-18 | pass→pass | 10,971 | 8,198 | -25% | 1 | 1 | 0% | 1,968 | 1,887 | -4% | 0 | 0 | — |
case-19 | pass→pass | 11,475 | 7,318 | -36% | 1 | 1 | 0% | 1,984 | 1,651 | -17% | 0 | 0 | — |
case-20 | pass→pass | 6,978 | 2,310 | -67% | 1 | 1 | 0% | 1,296 | 870 | -33% | 0 | 0 | — |
case-21 | fail→fail | 16,876 | 14,997 | -11% | 1 | 1 | 0% | 2,895 | 3,095 | +7% | 0 | 0 | — |
case-22 | pass→pass | 3,895 | 2,543 | -35% | 1 | 1 | 0% | 582 | 852 | +46% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +5 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.