What Claude Agent Skills mean for teams
Anthropic donated Agent Skills to the Agentic AI Foundation. What Claude Agent Skills mean for teams that share one skill across several AI tools.
Ideas on governing AI capability — skills, prompts, agents, MCPs, and secrets.
Anthropic donated Agent Skills to the Agentic AI Foundation. What Claude Agent Skills mean for teams that share one skill across several AI tools.
The LiteLLM compromise began with one key that was never revoked. What software supply chain security means once AI agents hold your credentials.
Custom GPTs are moving to ChatGPT plugins. Here is what transfers, what does not, and how to keep your instructions in one portable source.
ChatGPT custom instructions are personal settings. Here is what to put in them, and how teams keep one shared, versioned source.
AI agents and MCP setups are non-human identities. Vault secrets, refer by name, and substitute only on outbound HTTP.
Secrets sprawl scatters API keys across chats, env files and agent configs. Store them in a vault and refer by name. Packages never carry values.
Agent credentials belong in a vault, not in prompts. Packages carry name hints. Controlled substitution is designed to keep values out of model context.
New starters should get approved prompts, skills and MCP servers, not five conflicting copies. Approved catalogues and sharing levels make AI onboarding repeatable.
Your IDP owns the software service catalogue. Joyst governs your AI assets. The two do complementary jobs, and Joyst is not a Backstage alternative.
Packaging defines what an agent includes: skills, MCPs, prompts and secrets by reference. One private package beats rebuilding the setup for every harness.