The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the ha (Harness Anything) command-line tool to perform administrative operations, including listing agents (ha agent list), inspecting configurations (ha agent inspect), validating schemas (ha agent validate), and installing new agent identities (ha agent install). These are standard operational tasks within the tool's ecosystem.
[INDIRECT_PROMPT_INJECTION]: The skill creates and manages agent identities that process instructions and prompts, which represents a functional attack surface for instruction-based injections.
Ingestion points: The skill reads existing agent configurations and new declaration files (e.g., harness/agents/<id>.json) to define agent behavior.
Boundary markers: Encourages logical separation between judgmental instructions, procedural skills, and per-task facts to maintain clear execution boundaries.
Capability inventory: The process includes capabilities to write to the agent identity store and execute dry-run prompt assemblies.
Sanitization: Implements schema validation via ha agent validate to ensure declarations are well-formed before deployment.

create-agent