The Agent Skills Directory

[PROMPT_INJECTION]: Analysis of the instructions confirms they are strictly focused on architectural enforcement and do not contain any bypass or override attempts.- [DATA_EXFILTRATION]: The skill restricts file access to the local project's configuration (".lattice/config.yaml") and reference directories. No network activity or credential harvesting patterns were detected.- [REMOTE_CODE_EXECUTION]: No remote script downloads, package installations, or dynamic code execution commands are present.- [INDIRECT_PROMPT_INJECTION]: The skill ingests content from workspace documentation to drive its validation logic. This is the intended functionality for architectural enforcement and is limited to reading project-specific files.

architecture