The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted CSV/TSV data provided by users, which serves as a potential ingestion point for indirect prompt injection attacks. However, this risk is significantly mitigated by a multi-layered validation workflow:
Ingestion Points: User-provided files are normalized and profiled (rules/inputs-file-folder-list.md, rules/workflow-profile-sample.md).
Boundary Markers: Extensive use of structured YAML output and mandatory validation gates (rules/workflow-fail-closed-on-unmapped-columns.md).
Capability Inventory: Bash (for infrahubctl), Write (for YAML emission), and Read (for schema discovery).
Sanitization: Strict type coercion (rules/mapping-value-coercion.md) and dropdown label normalization (rules/mapping-dropdown-label-to-name.md) ensure that input data is treated as values rather than instructions.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute infrahubctl commands for branch management, data validation, and object loading. These operations are core to the skill's primary function. Security is maintained by enforcing a 'branch-first' principle, ensuring that no data is written directly to the default branch and that a full rollback is possible by simply deleting the temporary branch (rules/workflow-branch-before-load.md).
[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file utilizes dynamic context injection (!command) to gather local project information, such as finding existing schema and object files. These injections are limited to benign find commands used for environmental awareness and do not incorporate unvalidated user input or perform network exfiltration.

infrahub-importing-data