Define third-party skill source policy and disabled-by-default trust model

[yacosta738]

Goal

Define how Corvus should support third-party skills without weakening the product trust model.

Product Intent

Third-party skills should be:

• supported,
• disabled by default,
• installable only from user-defined sources,
• clearly separated from official skills in trust level and UX.

Current Context

Current remote skill installation is permissive and repository-based. That is too weak for the product model we want.

Questions to Close

• What counts as an allowed third-party source?
• How does a user explicitly opt in to third-party skill sources?
• Should sources be allowlisted in config, per workspace, or globally?
• What warnings, prompts, or approvals are required before install?
• How should Corvus represent trust level for installed skills?

Acceptance Criteria

• Third-party source policy is explicitly defined.
• Disabled-by-default behavior is explicit.
• Trust and UX boundaries between official and third-party skills are clear.
• Follow-up implementation issues can be created cleanly.

[linear]

DALLAY-168