700 Skills, 0 Governance: The Missing Layer in Agent Skill Stacks

Today's trending GitHub repos illustrate a growing trend: agent capabilities are rapidly stacking up, but the governance layer is lagging behind. Three repositories in particular stand out: calesthio/OpenMontage, a comprehensive agentic video production system; msitarzewski/agency-agents, an AI agency with various specialized agents; and earthtojake/text-to-cad, which adds CAD capabilities to agents.

The common thread is that developers are quickly amassing more skills and agents within their stacks, resulting in increasing capability density.

However, a crucial issue remains unaddressed: there is currently no governance framework in place to manage this growing complexity. Picture a team of 300+ skill files and a dozen specialized agents running in production. If a pipeline were to break, how would you determine who modified a skill, which version was running before the failure, who had permission to invoke the skill, and what the full audit trail of the past 24 hours looks like?

Current methods involve sifting through git logs, chat histories, and relying on memory, which is not the same as governance. Tools like thedotmack/claude-mem capture context across sessions but do not provide the same level of auditability. Memory helps agents avoid forgetting, while governance ensures the team does not burn down. The missing piece is a skill registry that treats skills as first-class artifacts with version management, role-based access control (RBAC), and comprehensive audit logs.

Iflytek/skillhub is an example of such a solution, offering a self-hosted, open-source skill registry designed for teams that have outgrown loose skill files. Paired with workflow orchestration platforms like iflytek/astron-agent, which handles multi-step orchestration with breakpoint recovery, governance at the skill level can extend to the workflow level. When a workflow fails on step 7, teams can simply fix the skill, republish through skillhub, and resume without having to restart from step 1.

In summary, the trending repos demonstrate that stacking capabilities works, but the next challenge is how to effectively govern a stack of 700 skills. Teams that prioritize governance will be better equipped to scale without incidents, while those that neglect it may find themselves spending their Mondays digging through chat logs instead.