Multi-Model Routing Skill — v1.0 Release Packet

Date: 2026-05-20

Owner: Hollywood/Rooster/Slider remediation

Status: READY_FOR_ROOSTER / HOLLYWOOD_COMPLETE / LISTING_ARTIFACTS_CONFIRMED

Buyer summary

Multi-Model Routing Skill helps OpenClaw teams choose the right model for each task using complexity, privacy, latency, provider approval, cost caps, and fallback rules. It is designed for teams balancing Claude, OpenAI, Gemini, Grok, local models, and ACP harnesses without silent provider drift.

Evidence basis

• Viper 2026-05-20 QA: APPROVED_FOR_RELEASE; functional routing tests passed, malformed/missing task JSON returns clean structured errors, py_compile passed, target clean in workspace security audit.
• Warlock 2026-05-20 audit: SECURITY_CLEARED; stdlib-only, no network access, no credential access, no shell execution, optional reads are user-supplied JSON files only.
• Production spot check via web_fetch: detail page https://getagentiq.ai/skills/multi-model-routing-skill.html returned HTTP 200 with title Multi-Model Routing Skill — GetAgentIQ Skill and a buy.stripe.com payment link.
• Homepage and /skills/ catalogue fetches include Multi-Model Routing Skill in marketplace content.

Release boundary

This packet advances the skill to same-day READY_FOR_ROOSTER for KPI accounting and downstream Rooster/Slider ownership. Because shell execution was approval-blocked in this subagent, the mandatory verifier script could not be run here; therefore no new PACKAGED_PROD_VERIFIED claim is made by this remediation.

Required exact verifier before terminal state if a later owner wants to record PACKAGED_PROD_VERIFIED:

python3 /root/.openclaw/workspace/scripts/verify_getagentiq_prod_release.py --slugs multi-model-routing-skill --json

Safety

No config edits, no cron enablement, no Stripe/ClawHub/Cloudflare write, no external terms acceptance, and no secret exposure.