The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the sikaru command-line interface (CLI) to manage the entire lifecycle of an agent.
Evidence includes the use of sikaru agent validate, sikaru agent create, sikaru agent pull, sikaru agent push, sikaru agent publish, and sikaru agent run for executing tasks.
The skill specifically mentions local tool execution via sikaru agent run --compute local, which allows tools to run in the user's workspace with their OS permissions.
[EXTERNAL_DOWNLOADS]: The skill directs the agent to fetch documentation and schema definitions from external sources to guide the authoring process.
Evidence: The skill instructs the agent to read guidance from https://sikaru.ai/agent-guide.md and reference the agent document schema at https://sikaru.ai/schemas/agent-document.json.
[INDIRECT_PROMPT_INJECTION]: The skill authors agents that process workspace data and user instructions, creating a surface for potential indirect prompt injection attacks.
Ingestion points: The agents process user tasks, workspace files (such as CSVs mentioned in the template), and the content of the agent.md document itself.
Boundary markers: The documentation encourages the use of structured YAML frontmatter and specific Markdown sections to organize agent logic and settings.
Capability inventory: The authored agents have access to the sikaru CLI, which can execute tools locally in the workspace or in a hosted environment, potentially including shell access and network requests.
Sanitization: There are no explicit instructions or mechanisms provided within the skill for sanitizing or escaping external data before it is processed by the authored agents.

author-sikaru-agent