docs(triage): allow authorized agent labeling

[ss-o]

Summary

• allow LLM agents to apply canonical labels when a maintainer explicitly authorizes the triage scope
• keep label changes draft-only without that authorization
• keep comments and issue closures human-only
• preserve the stricter draft-only boundary for unattended recurring reviews

Instruction impact review

1. Classification: This is scoped operational policy for interactive issue and pull-request triage.
2. Consumers and contexts: All supported agent runtimes performing the triage task across z-shell repositories consume this runbook through the public instruction manifest.
3. Canonical owner: runbooks/triage.md remains the correct canonical owner.
4. Duplication or contradiction: No duplicate owner is introduced. runbooks/recurring-operations.md intentionally remains stricter for unattended and recurring reviews.
5. Manifest impact: No manifest route needs to change because the existing required runbook-triage surface and triage task routing remain correct.
6. Runtime delivery: Every supported runtime continues to receive the rule through the required public manifest route; no optional hook, skill, or adapter becomes authoritative.
7. Generated output and limits: The public baseline and private overlay are unchanged, so no generated AGENTS.md update is part of this repository change. Private meta-workspace gitlink and composite reconciliation remains a separate post-merge change.

Validation

• python3 scripts/validate-agent-policy.py: passed
• python3 -m unittest scripts/test_validate_agent_policy.py -v: 75 tests passed
• git diff --check: passed

No issue comments, closures, labels, repository settings, or automation behavior are changed by this pull request.