The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external deliverables for presentation to the user. While this creates a surface for indirect prompt injection, the skill mitigates this by requiring explicit user approval (APPROVED/REJECTED) and providing clear boundary markers in the presentation template. The severity is low as it defines a safety gate rather than bypassing one.
[DATA_EXPOSURE_&_EXFILTRATION]: The skill logs approval and rejection metadata to local files (~/.ai-software-agency/audit.log and tasks.md). This is a standard project management practice and does not involve exfiltration of sensitive information or credentials to external domains.
[COMMAND_EXECUTION]: The skill references a memory-sync skill at the end of the session. As the implementation of memory-sync is not provided, this is noted as a standard inter-skill call pattern for persistent memory management.

governance-gate