Feature request: install-time verification hook for plugins/skills; content re-validation for marketplace caches

Problem

Empirical findings against 2.1.210 (full method: https://github.com/mindmodelai/contextlock/blob/master/docs/claude-code-surface.md):

• Installed plugin cache content is never re-validated after install.
• For local-source marketplaces the cache is bypassed entirely - content is served live from the source directory, so anything with write access to that directory changes what runs with no check.
• Marketplace entries pin git SHAs at best; CLAUDE.md, skills, and rules have no signing or install-time validation hook anywhere in the pipeline.

Proposal (two independent pieces)

1. An install-time verification hook: an operator-configured command that runs after a plugin/skill is staged but before it is installed, receiving the staged path on stdin and returning allow/block - failing closed. OpenClaw ships exactly this (security.installPolicy) and it has proven to be the single most useful integration point for external integrity tooling, because it enables verification without the host tool adopting any cryptography or policy semantics itself.

1. Longer term: signed plugin/skill artifacts with client-side verification - the OWASP AST01 recommendation, and the direction mainstream ecosystems already took (VS Code Marketplace extension signing, npm provenance attestations).

These compose: (1) is small and unblocks the ecosystem now; (2) is the durable fix.

---
Context: ContextLock (Apache-2.0, https://github.com/mindmodelai/contextlock) implements DSSE-signed manifests + Sigstore keyless verification for instruction files and would plug into (1) unchanged. Related ask: #77988 covers the load-time side. Drafted with AI assistance, human-reviewed.