Agent Skills moved to a new foundation — what changes for teams building on it?

October 30, 2026 · 4 min read

SentX SI

On October 22, 2026, Anthropic announced that it is donating Agent Skills — the open standard that gives AI agents specialized knowledge and workflows, published in December 2025 after an October 2025 launch — to the Agentic AI Foundation (AAIF), a body under the Linux Foundation. For a team that has wired multi-tool agent workflows to the standard, the honest summary is this: your access does not change, the home does, and the guarantees you can plan against are narrower than the announcement's framing suggests. The post states a neutral home, named maintainers, and an open contribution path. What sits beyond that splits two ways: versioning and security guidance arrive only as proposals, while deprecation and dispute resolution are not addressed in the post at all.

Source: Donating Agent Skills to the Agentic AI Foundation

What the announcement commits to

The donated package is specific: the Agent Skills name and logo, the GitHub organization, and the agentskills.io domain. Stewardship moves to AAIF, which Anthropic says it co-founded with Block and OpenAI in 2025 and to which it had already donated the Model Context Protocol (MCP). Anthropic states that AAIF will provide a neutral home for the standard going forward, ensuring the community has a say in where it goes next. Maintenance stays with people rather than a single company: engineers from Anthropic, Google, OpenAI, and other companies will maintain the standard as individuals, with David Soria Parra and Mahesh Murag named as Lead Maintainers. You can use the standard or contribute to it without joining AAIF, and the project says it will continue to prioritize community input and transparent decision-making, as with MCP. Anthropic also reports adoption numbers worth carrying their source with them: more than 70 agent platforms — including ChatGPT, Cursor, GitHub Copilot, and VS Code — have integrated the standard since December 2025, with 1 million skills published and nearly 280 million installed. Those are Anthropic's figures, not independently verified ones.

Source: Donating Agent Skills to the Agentic AI Foundation

What it leaves open

The roadmap — versioned releases, a formal proposal process, and security guidance for people who write and install skills — is described as proposed. Read that word carefully, because it defines the boundary of what you can rely on today. The announcement does not specify a versioning policy, so there is no stated rule for how releases are numbered or whether a new version preserves compatibility with the one your loaders implement. There is no deprecation process and no release cadence. Security guidance is named but undelivered: no review gate describes who vets a community skill before it reaches your machines. Dispute resolution is absent — the post does not say what happens when maintainers disagree, and the proposal process that would govern specification changes is itself part of the proposal. "Other companies" appears once in the maintainer sentence and is never enumerated. None of this is a criticism; it is the shape of a standard early in its new home. But for a production dependency, each missing item is a decision your team currently makes on its own.

Source: Donating Agent Skills to the Agentic AI Foundation

A worked example: mapping a multi-tool workflow

Consider a team that ships one document-review skill — a SKILL.md folder with helper scripts — across three surfaces: an internal coding assistant, a commercial IDE, and a batch pipeline that also ingests community-published skills. Four dependencies matter, and the announcement covers them unevenly. First, continued access: the standard remains usable without membership, and because the GitHub organization and the agentskills.io domain are themselves the donated assets, the spec and repository keep their addresses. Second, compatibility: the pipeline assumes the next spec revision will not break what it parses. That assumption is not covered — versioned releases are proposed, and no compatibility policy is stated — so the team pins the spec version its loaders implement and re-verifies when a release process actually lands. Third, trust: the pipeline installs community skills. With no security gate named, the team keeps its own vetting step and treats future guidance as input, not protection. Fourth, contribution: when the team fixes a bug upstream, the open contribution path and the project Discord — the stated channel for feedback on proposed specification changes — remain available without any membership. This mapping is illustrative, not a record of any deployment; its point is structural. Where the announcement states a fact, the team inherits it. Where the announcement is silent, the team's own controls fill the gap.

Source: Donating Agent Skills to the Agentic AI Foundation

Verify the state, not the announcement

The durable change for adopters is therefore a habit rather than a migration: the standard now has a neutral home and named individuals maintaining it, but its operational guarantees are still forming, and the announcement is a snapshot, not a warranty. SentX SI builds its own documentation around the same split. Its model card labels itself a first-party product disclosure rather than an independent certification, and its checkpoint builder carries the note that it formats your notes; it does not verify them. Structure and provenance are provided; checking is left to you — which is exactly the position a team is in with a newly donated standard. The practical checklist is short: read the current specification at agentskills.io, watch the GitHub organization for the release and proposal processes taking shape, and join the Discord if your team wants a seat in the decisions still being made.

Source: Donating Agent Skills to the Agentic AI Foundation; Victoria: public model card