Skip to content

gracefullight/pkgs / .agents/skills/oma-tf-infra

oma-tf-infra

Verified

Create or review Terraform infrastructure and plans. Use for cloud resources, IAM, networking, state management, and infrastructure changes.

resources/iso-42001-infra.md

View on GitHub
# TF Infra Agent - ISO/IEC 42001 / ISO 22301 / ISO/IEC/IEEE 42010 Guide

Use this file when the task involves AI systems and the user wants infrastructure controls or audit-friendly recommendations aligned with ISO/IEC 42001, continuity and recovery guidance aligned with ISO 22301, or infrastructure architecture documentation guidance aligned with ISO/IEC/IEEE 42010.

## Scope

This guide is for:

- **infrastructure controls** that support AI governance
- **continuity and recovery controls** for infrastructure resilience
- **architecture description guidance** for infrastructure decisions

It does **not** replace:

- organizational policy
- model governance
- human oversight design
- AI risk management process ownership

Those belong primarily to PM, QA, backend, or security governance work.

## What TF Infra Agent Should Cover

For AI systems, focus on infrastructure controls such as:

- IAM and least privilege
- service-to-service identity
- network segmentation and private connectivity
- encryption at rest and in transit
- secrets management
- centralized logging and audit trails
- monitoring and alerting
- backup and retention
- environment isolation
- change traceability in CI/CD and Terraform state
- continuity and failover design
- architecture views, dependencies, and decision rationale

## Recommended Control Areas

### 1. Access Control

Implement:

- least-privilege IAM roles
- workload identity / OIDC for CI/CD
- separation of admin, deployer, runtime, and read-only roles
- scoped access to model endpoints, vector stores, object stores, and databases

Avoid:

- shared admin roles
- long-lived keys
- wildcard permissions to AI-related data or services

### 2. Network Boundaries

Implement:

- private subnets for sensitive workloads
- VPC/VNet/VCN peering or private service endpoints
- restricted egress where feasible
- ingress limited to approved paths such as ALB, API Gateway, or internal load balancers

Use stronger isolation when:

- prompts or training data contain sensitive data
- model outputs affect business-critical actions
- cross-region or cross-tenant exposure is a concern

### 3. Encryption and Secrets

Implement:

- provider-managed KMS or customer-managed keys where required
- encrypted storage for object stores, DBs, queues, and state backends
- TLS for internal and external traffic where supported
- secrets in cloud secret managers, not in Terraform variables committed to git

### 4. Logging and Auditability

Implement:

- centralized audit logs for IAM, network, storage, and runtime services
- immutable or retention-locked log targets where needed
- request tracing across gateways, application services, and data stores
- Terraform plan/apply evidence in CI/CD logs

For AI-related systems, especially capture evidence around:

- who changed infrastructure
- who can access model-serving paths
- where prompts, embeddings, or outputs are stored
- whether logging retention matches policy

### 5. Monitoring and Alerting

Implement:

- uptime and health monitoring
- resource utilization metrics
- security event alerting
- data egress anomaly alerting where available
- alerting for failed backups, replication, or logging pipelines

If the AI system is production-critical, recommend:

- SLO-aligned alerting
- model endpoint availability monitoring
- queue/backlog monitoring for ingestion pipelines
- vector or search cluster health checks if present

### 6. Retention, Backup, and Recovery

Implement:

- backup policies for stateful systems
- snapshot retention and lifecycle rules
- archival tiers for logs and historical data
- restore procedures and periodic recovery validation

For AI systems, think about:

- prompt / output retention requirements
- embedding index rebuild strategy
- backup of canonical data stores rather than only derived indexes

### 7. Environment and Tenant Isolation

Implement:

- separate environments by account/project/subscription/compartment when practical
- isolated state backends per environment
- tagging/labeling for ownership and cost attribution
- tenant-aware segmentation when AI data is multi-tenant

## 8. ISO 22301 Continuity View

For infrastructure continuity, recommend improvements around:

- critical service identification
- RTO / RPO assumptions
- backup scope and retention
- restore verification
- redundancy and failover
- cross-region or cross-zone strategy
- dependency mapping
- manual fallback and degraded-mode assumptions

Examples of infra-focused suggestions:

- multi-AZ or multi-zone deployment for critical paths
- warm standby or replicated storage for essential services
- tested snapshot restore workflow
- documented dependency chain for identity, DNS, storage, and network services
- alerting for failed replication or backup jobs

## 9. ISO/IEC/IEEE 42010 Architecture View

Use 42010 as an architecture description guide, not as a control checklist.

Capture:

- stakeholders
- concerns
- architecture views
- interfaces and dependencies
- constraints
- rationale and tradeoffs

For infrastructure work, useful views include:

- context view
- deployment / runtime view
- network boundary view
- identity and trust view
- resilience / recovery view
- observability view

Useful documentation prompts:

- Who needs this architecture description?
- What concerns are they trying to answer?
- Which infrastructure elements and interfaces matter?
- What tradeoffs were chosen and why?
- What dependencies create operational risk?

## Review Questions

When applying ISO/IEC 42001 / ISO 22301 / ISO/IEC/IEEE 42010 thinking, ask:

- Can access to AI-related data paths be audited?
- Are model-serving and data-serving paths protected with least privilege?
- Are logs centralized, retained, and protected against tampering?
- Are backups, retention, and restore procedures defined?
- Are continuity objectives such as RTO / RPO explicit?
- Are failover paths and critical dependencies visible?
- Is network exposure minimized?
- Is CI/CD access ephemeral and traceable?
- Is there clear ownership for each critical resource?
- Are the stakeholders, concerns, and architecture rationale documented clearly?

## Output Pattern

When relevant, add a short section like:

```md
## ISO/IEC 42001 / ISO 22301 / 42010 Notes
- Access control:
- Network boundary:
- Encryption:
- Logging / auditability:
- Monitoring / alerting:
- Backup / retention:
- Continuity / RTO / RPO:
- Failover / dependency view:
- Stakeholders / concerns / views:
- Gaps:
- Recommended Terraform changes:
```

## Guardrails

- Do not claim ISO/IEC 42001 compliance from Terraform alone
- Do not claim ISO 22301 compliance from backup settings alone
- Do not expand this skill into full AI governance process design
- Do not turn 42010 into paperwork without architectural value
- Keep suggestions concrete and infra-implementable
- If the gap is organizational rather than infrastructural, say so explicitly
writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 322kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall