At https://claude.ai/directory/manage/new/plugin, the Compliance step requires confirmation that a plugin does not execute code outside its declared MCP servers. How does that declaration apply to a skill that instructs Claude to run a bundled local Python validator through the host's normal execution tools?
UX3 Product Design Harness contains one skill with three local Python helpers and no MCP servers or hooks. Its skill instructions require the canonical validator, so describing that execution as occurring inside a declared MCP server would be inaccurate. We have not checked the declaration or submitted the plugin.
The official Skills overview documents bundled scripts executed through bash. The directory pre-submission checklist also distinguishes scripts that a skill tells Claude to run from scripts followed by particular scanner checks. We could not find a directory-specific definition explaining how those documented capabilities fit this required declaration.
Could you clarify its scope or document the supported submission route for a complete skill bundle with local validators? We are not requesting a review bypass or proposing a fictitious MCP wrapper.
References:
- https://claude.com/docs/plugins/submit
- https://claude.com/docs/plugins/pre-submission-checklist
- https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview
- https://support.claude.com/en/articles/13145358-anthropic-software-directory-policy
- Public source and privacy disclosure: https://github.com/cis2042/ux3-product-design-harness/tree/7f9f724c5e31813d1c35767f9f454fe9a1e96891
Portal Source validation passed for abf458da5d3915a40690f71d535f90fd6f05425c. Two UNREAD_ASSET_REFERENCED findings remain held for review because brand documentation and a test refer to a JPEG. Those findings are separate from this declaration question; we have not attempted to suppress them. The later commit linked above changes only the README's privacy/data-handling disclosure.
Observed on 2026-10-23. Native Claude Code 2.1.309 component inventory exposes one skill and zero agents, hooks, MCP or LSP servers. This report concerns the portal declaration wording, not a claim that Anthropic prohibits all local skill scripts.
Solaris Dev Shop has the same declaration question before a purchase made solely to submit. Our current Free account stops at the paid-plan gate, so we have not observed or accepted this declaration ourselves; the wording reported above needs clarification for our use case too.
Our exact prepared0.68 Claude package includes61 specialist workflows plus Chief of Staff and executable local intake, evidence and recovery helpers. It declares no MCP servers or hooks. Its README explicitly discloses that a selected workflow can ask the host to run local commands; that behavior is central to the product, so we cannot truthfully describe every execution as inside a declared MCP server. Local strict manifest validation and source CI passing do not resolve this compliance wording.
The directory checklist distinguishes scripts that a SKILL.md tells Claude to run from automatically followed commands; the Skills overview also documents bundled code and local execution.
Could a maintainer clarify whether the compliance declaration covers all explicit user-requested host execution, or only execution initiated outside the documented skill/tool flow? If the former, what supported submission route applies to a complete skills bundle with local validators/helpers? We are not proposing a fictitious MCP wrapper or asking to omit capabilities, bypass review, or certify a declaration we cannot meet. Joining this existing technical question rather than opening a duplicate issue.