The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted pipeline YAML and possesses the capability to modify it, which creates a potential surface for indirect prompt injection.\n
Ingestion points: The skill fetches pipeline YAML from the Harness platform using the harness_get tool (Phase 1) and lists policy sets via harness_list (Phase 9).\n
Boundary markers: The instructions do not define explicit delimiters or 'ignore' instructions for the processed YAML content to prevent embedded instructions from influencing agent behavior.\n
Capability inventory: The skill is equipped to update the pipeline using the harness_update tool, which can rewrite the pipeline logic.\n
Sanitization: No specific sanitization or escaping mechanisms for external YAML data are mentioned in the configuration logic.\n
Mitigation: The skill incorporates a mandatory user confirmation turn (Phase 10) before applying changes via harness_update, providing a manual checkpoint to verify the generated YAML.

enforce-sbom