Auto-route tasks to the cheapest Claude model that works correctly. Three-tier progression: Haiku → Sonnet → Opus. Classify before responding. HAIKU (default): factual Q&A, greetings, reminders, status checks, lookups, simple file ops, heartbeats, casual chat, 1-2 sentence tasks. ESCALATE TO SONNET: code >10 lines, analysis, comparisons, planning, reports, multi-step reasoning, tables, long writing >3 paragraphs, summarization, research synthesis, most user conversations. ESCALATE TO OPUS: archi
Security Analysis
high confidenceThe skill is an instruction-only routing policy that consistently matches its stated purpose (choosing among three Claude models) and requests no extra permissions, installs, or credentials.
Name/description match the runtime instructions: the SKILL.md contains routing rules, examples, and decision logic for choosing Haiku/Sonnet/Opus. There are no unrelated environment variables, binaries, or install steps requested that would contradict the stated purpose.
The instructions are narrowly scoped to classification and escalation rules and include example calls (sessions_spawn, cron payload) appropriate to an agent that can spawn model sessions. The doc does not instruct reading unrelated system files, secret env vars, or exfiltrating data to external endpoints.
No install spec and no code files are present (instruction-only). That minimizes on-disk footprint and there is no download/execute behavior to evaluate.
The skill declares no required environment variables, credentials, or config paths. The decision rules do not reference any hidden or unrelated credentials, so requested privileges are proportionate to a routing policy.
always:false and default autonomous invocation are appropriate for this kind of policy. The skill does not request persistent system-wide changes or modifications to other skills' configs.
Guidance
This skill is coherent and low-risk from a permissions perspective because it's only a policy document with examples — it asks for nothing sensitive. Before installing: (1) confirm your agent platform enforces billing/usage limits (cheaper models may lower cost but also capability), (2) test the 30-second escalation heuristic on representative tasks to ensure it doesn't under- or over-escalate, and (3) avoid routing sensitive or high-stakes decisions to lower-tier models (Haiku) — explicitly whitelist such tasks to Sonnet/Opus. If you need stronger guarantees, request an implementation that logs routing decisions and includes safeguards for privacy-sensitive inputs.
Latest Release
v1.0.0
Initial release - Three-tier Claude routing for cost optimization
Popular Skills
Published by @millibus on ClawHub
