Core Usage
agent-bom-discover-aws is a Python-based discovery skill that inventories AWS AI infrastructure—specifically Bedrock Agents, SageMaker endpoints, Lambda functions, ECS tasks, EKS workloads, Step Functions, and EC2 instances—to generate a canonical "agent-bom" JSON artifact. It operates entirely within the operator's local environment, leveraging the existing AWS SDK credential chain (SSO, WebIdentity, or STS assumed roles) to make read-only API calls. The skill supports three modes: discover-only (emit inventory and stop), scan-local (hand off to agent-bom scanner), and export (write SARIF/JSON). The default discover-only mode ensures zero external data exposure unless explicitly requested.
Key Strengths
- Credential Hygiene: Enforces short-lived credentials; explicitly prohibits accepting pasted access keys or printing credential values. Credentials never leave the operator environment.
- Minimal Blast Radius: Discovery-only by default; no modifications to AWS resources. Optional scan/export requires operator approval.
- Evidence Integrity: Emits schema-validated JSON with provenance metadata (
source_type: skill_invoked_pull,observed_via: aws_sdk), enabling audit trails. - Scope Control: Granular flags (
--include-ecs,--include-lambda, etc.) allow operators to start narrow and expand deliberately. - Open Source: Apache-2.0 licensed with public GitHub repository and PyPI distribution, enabling community audit.
Limitations & Risks
- Dependency on Operator IAM: Requires correctly configured read-only IAM policies; misconfiguration could expose sensitive metadata or allow unintended actions.
- Python 3.11+ Requirement: Legacy environments may need upgrades.
- Regional Coverage: Must specify regions explicitly; multi-region deployments require iterative execution.
- No Real-time Monitoring: One-time inventory snapshot; drift detection requires scheduled re-runs.
- Data Redaction: While credential patterns are redacted, operators must verify output sanitization before sharing.
Suitable Users
Cloud security engineers, AI governance teams, and compliance auditors managing AWS-native AI workloads who need agent/workload visibility without granting third-party services persistent cloud access.
Security Assessment
| Dimension | Rating | Rationale |
|-----------|--------|-----------|
| Credential handling | Strong | Short-lived STS/SSO only; no long-lived keys; no credential transmission to agent-bom |
| Data flow | Controlled | Local SDK calls → local JSON; optional scan requires explicit operator command |
| Privilege escalation | None | Read-only IAM scope; no resource modification |
| Telemetry | None | Explicit `telemetry: false` in manifest |
| Persistence | None | `persistence: false`; operator controls output path |
Residual Risk: If executed with overly permissive IAM policies, the skill could expose sensitive resource metadata. Operators must validate IAM scope via agent-bom trust before execution.