The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill manages agent permissions, including an unrestricted field for full access, and identity switching via host.agents.switch. These management capabilities are restricted by design through mandatory user confirmation and external approval mechanisms described in the instructions.
[INDIRECT_PROMPT_INJECTION]: The skill allows users to define the systemPrompt for specialist agents, creating a surface where instructions could be injected into new agent identities.
Ingestion points: User-supplied strings for systemPrompt, displayName, and description are ingested into the host.agents.create and host.agents.update SDK calls in SKILL.md.
Boundary markers: The instructions mandate a prefix for the generated identity: "You are {displayName}."
Capability inventory: The skill utilizes the host.agents SDK to create, update, and delete agent specialists and modify their access levels.
Sanitization: Error reporting is explicitly configured to remove system instructions, credentials, headers, and tokens before presentation to the user to prevent data exposure.

customize