The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates the creation of new agent instructions based on user-provided templates, which represents a potential surface for indirect prompt injection.
Ingestion points: Data provided to the prompt templates, specifically the AGENT_NAME, DESCRIPTION, and SYSTEM_PROMPT fields as described in the generation workflow.
Boundary markers: The generation logic in agent_generator.py does not include specific boundary markers or instructions to the LLM to ignore instructions embedded within the user-provided data during the assembly of the generated agent files.
Capability inventory: The generated agent files are intended to be installed in the .claude/agents/ directory, allowing them to persistently influence the agent's behavior and access tools within the project environment.
Sanitization: The AgentGenerator script validates that agent names follow kebab-case formatting but does not perform sanitization, filtering, or escaping of the content provided for the system prompt or description fields.

agent-factory