Security workflows fail on fork PRs at the report-comment step (403), skipping SkillSpector's Enforce gate

On a PR from a fork, GITHUB_TOKEN is read-only, so the report-comment step in both security workflows fails with 403 Resource not accessible by integration:

• Component Security Validation → Comment PR with Results
• Skill Security Scan → Comment PR with results

Every step before it completes, and the job goes red at the comment step. Recent examples:

PR	Job	Run	Scan result	Job conclusion
#773	SkillSpector (changed skills)	30843950116	Scanned=1 flagged=0 errors=0	failure (403 at comment step)
#773	Security Audit	30843949984	audit ran (exit masked by continue-on-error)	failure (403 at comment step)
#770	SkillSpector (changed skills)	30724621083	—	failure; PR merged anyway
#770	Security Audit	30724621075	—	failure; PR merged anyway

Two effects beyond the cosmetic red X:

1. SkillSpector's Enforce gate never runs on fork PRs. It is ordered after the comment step, and the 403 fails the job first (run 30843950116 shows it as skipped). The HIGH/CRITICAL block only enforces on same-repo PRs, and for outside contributors every PR is a fork PR.
2. Red has stopped meaning "flagged." Every fork PR gets the same two red X's whether its scans passed or not, so a flagged skill produces the same signal as a clean one; #770 merged over them. On the Security Audit side the audit step is continue-on-error: true, so the comment 403 is that job's only failure mode: red there never means the audit failed. (The CI-mode audit currently exits non-zero on every run, with 412 of 769 components failing the strict checks repo-wide, and even that never reds the job.)

The minimal fix is to skip the comment step when the PR is from a fork (github.event.pull_request.head.repo.full_name == github.repository). Fork contributors get no comment today either (the step 403s before posting), and the reports stay available as artifacts. With the step skipped instead of failing, the jobs go green when scans pass and Enforce gate actually runs.

A fuller fix would move the commenting into a workflow_run job, which runs with write permissions and could post the report on fork PRs too. That is a bigger change; the guard works either way as a first step.