The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes patterns for ingesting untrusted data from external files, which can influence agent actions or deployment parameters.
Ingestion points: The skill uses fs.readFileSync("./deploy-config.json", "utf-8") in SKILL.md to load deployment configurations at runtime.
Boundary markers: The provided implementation lacks explicit validation or boundary markers for the data extracted from the configuration file.
Capability inventory: The skill utilizes high-privilege capabilities including stack.up(), stack.destroy(), and stack.setConfig() to manage cloud infrastructure.
Sanitization: No specific sanitization or schema validation beyond standard JSON parsing is demonstrated for the external configuration input.
[COMMAND_EXECUTION]: The skill demonstrates how to programmatically execute significant infrastructure lifecycle commands via the Pulumi engine.
The documentation provides patterns for calling stack.up() and stack.destroy(), which trigger the execution of infrastructure-as-code programs to provision or tear down resources.
[DYNAMIC_EXECUTION]: The skill promotes the use of the "Inline Source" pattern, which involves executing infrastructure programs defined as runtime functions.
The program function property within LocalWorkspace.createOrSelectStack allows for the definition and execution of infrastructure logic dynamically during the orchestrator's execution flow.

pulumi-automation-api