The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes existing agent configurations and task data to draft new agents. While this creates a surface for indirect instructions, the risk is mitigated by a mandatory human-in-the-loop verification step.
Ingestion points: existing agent files in skills/specialists/ and task metadata in ~/.ai-software-agency/tasks.md.
Boundary markers: The draft is enclosed in a structured 'Governance Gate' proposal.
Capability inventory: The skill uses Write and Edit tools to persist approved agents.
Sanitization: A mandatory human approval process via the governance-gate skill is required for all file writes.
[DATA_EXPOSURE]: The skill accesses application-specific state and logs located in ~/.ai-software-agency/. This access is necessary for its primary function of agent lifecycle management and audit logging.

hire-agent