setting-up-continuum

Pass

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • Secure Credential Management: The skill enforces a safety protocol for AWS CLI operations, requiring the agent to verify the account and region via STS before performing any actions and mandating explicit user approval for every write operation. It also instructs the agent to ignore existing endpoint configurations to prevent unintended network routing.
  • Cryptographic ID Generation: A local Node.js script is used to generate 256-bit random hex characters for AWS IAM external IDs. This follows security best practices by using the Node.js CSPRNG (node:crypto) rather than user-provided or guessable strings, which mitigates the risk of IAM role assumption attacks.
  • Local Data Isolation: Deployment configuration is stored in a dedicated local file (.continuum/deployment-profile.local.md) and automatically added to the local Git ignore list (.git/info/exclude). This approach keeps sensitive environment topology facts on the developer's machine and prevents them from being accidentally committed to version control.
  • Defensive Resource Creation: When creating AWS resources like S3 buckets, the skill uses defensive flags such as --expected-bucket-owner and --object-ownership BucketOwnerEnforced to prevent common misconfigurations and ensure the agent interacts with the correct infrastructure.
  • Quota Preflight Validation: The skill includes instructions to manually calculate and verify repository counts against documented AWS service quotas before initiating environment creation, which prevents partial or failed deployments due to account limits.
  • Untrusted Data Handling: The skill processes repository content and AWS API responses, which are potential ingestion points for untrusted data.
  • Ingestion points: Source code, environment configurations, and AWS service outputs.
  • Boundary markers: Uses the 'shared AWS safety prelude' to separate context.
  • Capability inventory: Performs AWS CLI write actions and local file system updates.
  • Sanitization: The instructions explicitly mandate treating all external data as data only, never as executable instructions or agent commands, which significantly reduces the risk of indirect prompt injection.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 19, 2026, 12:39 AM