Add reusable SkillSpector skill scanning

[KristianGrafana]

What this adds

• Add a reusable workflow that runs pinned SkillSpector v2.9.6 directly.
• Scan every nested SKILL.md with --no-llm.
• Use trusted base-tree baselines for pull requests.
• Report only new HIGH or CRITICAL findings with confidence of at least 0.8.
• Update one stable pull request comment and upload a sanitized JSON report.

Value

Repositories get one copy-and-paste workflow and one CI check before merge. The narrow default policy keeps existing, low-severity, and low-confidence findings out of pull request comments.

Security and privacy

• Pin SkillSpector and all external actions to full commit SHAs.
• Reject incomplete scans, symlinks, Git submodules, and unsupported scope exclusions.
• Use only the trusted base checkout to create SkillSpector baselines.
• Omit scanner evidence, code snippets, and file content from comments and artifacts.
• Do not send skill file content to an LLM. SkillSpector can send declared package coordinates to OSV.dev.

Validation

• Run the exact workflow shell against a safe skill and a real CRITICAL SkillSpector fixture.
• Confirm that a trusted base baseline suppresses an existing finding.
• Confirm that hidden unsupported content, symlinks, submodules, and partial scans fail closed.
• Pass actionlint, zizmor --pedantic, YAML parsing, Bash parsing, and git diff --check.
• Complete independent security, runtime, and documentation reviews.

Release note

The usage example pins signed implementation commit 8d0e6780f8cbb5e73ffec054615082ca2a26c536. If the merge strategy rewrites this commit, update the example to a stable signed commit before the feature branch is deleted.

[isaiah-grafana]

LGTM