The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructs the agent to "request an unsandboxed retry" if commands are blocked by security restrictions in the execution environment (SKILL.md). This is an attempt to influence the agent to bypass security controls.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external OpenAPI documents to model Konnect resources, creating a vulnerability surface for malicious content in specs to influence the configuration pipeline.
Ingestion points: OpenAPI specification files referenced via !file tags in references/apiops-openapi.md.
Boundary markers: None provided to isolate the extracted content.
Capability inventory: Execution of state-mutating kongctl commands (apply, sync, delete) and repository scripts.
Sanitization: Content from specs is interpolated directly into declarative manifests without validation.
[DATA_EXFILTRATION]: Troubleshooting procedures in references/troubleshooting.md direct the agent to inspect environment variables (env | grep '^KONGCTL_') and organizational metadata, which can lead to the exposure of sensitive authentication tokens.
[COMMAND_EXECUTION]: The skill facilitates the execution of the kongctl CLI and arbitrary repository scripts (e.g., ./scripts/konnect-sync.sh) to synchronize environment state.
[EXTERNAL_DOWNLOADS]: The skill recommends using official GitHub Actions such as kong/setup-kongctl and kong/setup-deck for CI/CD workflows. These are documented as vendor-provided utilities.

kongctl-declarative