Define and enforce a skill capability manifest

[Scriptception]

Problem

Skills can include instructions, scripts, references, and assets, but metadata does not declare filesystem, network, command, credential, or external-tool expectations for security review and installation.

User outcome

Users and reviewers can understand a skill's required capabilities before installation or execution.

Scope

• Define a versioned capability schema for files, commands, network, tools, credentials, and generated artifacts.
• Default undeclared capabilities to denied or manual review.
• Validate paths, symlinks, executable content, and referenced files.
• Render a human-readable capability summary before install.

Non-goals

• Sandboxing every supported agent.
• Claiming metadata can make malicious instructions safe.

Acceptance criteria

• Every official skill has an explicit capability declaration.
• Validation rejects path traversal, missing assets, undeclared executables, and malformed capabilities.
• Install preview shows capability and trust information.
• Adapter output preserves or links to required warnings.

Validation

• Test benign and adversarial skill bundles including symlinks and undeclared scripts.
• Review official skills under the new schema before release.

Relationships

• Feeds authenticated catalog trust decisions.
• Complements provenance and evaluation evidence.