The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several command-line interface (CLI) tools to manage infrastructure and verify the execution environment.
In SKILL.md, aws sts get-caller-identity is executed to ensure the agent is operating within the correct AWS account and role.
In SKILL.md, terraform init, terraform plan, and terraform show are used to initialize the project and generate the JSON-formatted plan for analysis.
In SKILL.md, terraform apply is executed to perform infrastructure changes, though this is guarded by a manual approval step.
[INDIRECT_PROMPT_INJECTION]: The skill processes data generated from the local environment which is then passed to subagents for analysis, creating a potential surface for indirect prompt injection.
Ingestion points: The content of plan.json, which reflects the local Terraform configuration and state, is interpolated directly into the prompts for the terraform-plan-analyzer and security-reviewer subagents in SKILL.md (Step 3).
Boundary markers: The prompts use descriptive headers like Plan JSON: to delimit the content, but lack specific instructions for the subagents to ignore potential command-like patterns or malicious instructions embedded within the infrastructure data.
Capability inventory: The skill has significant capabilities, including the ability to modify cloud infrastructure via terraform apply and execute arbitrary CLI commands in the local shell environment.
Sanitization: There is no explicit sanitization, filtering, or schema validation performed on the plan.json content before it is presented to the analysis agents.

terraform-plan-review