Save Money 4.2.0
v1.0.0Auto-detect task complexity for Claude models (Haiku + Sonnet). Route simple tasks to Haiku, escalate complex ones to Sonnet. Save 50%+ on API costs. | 適用 Cl...
⭐ 0· 294·4 current·4 all-time
by@sieyer
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
The name/description promise (auto-classify and route between Haiku and Sonnet) matches the SKILL.md: it contains explicit classification rules and a single escalation call pattern. No unrelated credentials, binaries, or config paths are requested.
Instruction Scope
Instructions require the agent to classify every incoming task and call sessions_spawn(...) with the full task when escalation triggers fire. This is coherent with the routing purpose, but it does mean the entirety of escalated task text will be sent to the Sonnet model — review whether that content may include sensitive data before allowing automatic escalation.
Install Mechanism
Instruction-only skill with no install spec and no code files. Low install risk because nothing is written to disk and no external packages or downloads are performed.
Credentials
The skill declares no environment variables, credentials, or config paths. The only outward action is an internal sessions_spawn() call specifying a Sonnet model, which aligns with the purpose.
Persistence & Privilege
always is false and the skill does not request persistent system changes or modify other skills. It requests normal autonomous invocation behavior (disable-model-invocation=false), which is expected for a routing skill.
Assessment
This skill is internally consistent with its stated goal of routing tasks between Haiku and Sonnet. Before installing: (1) confirm you are comfortable with the agent automatically sending full task text to the Sonnet model (cost and data exposure), (2) verify your platform's sessions_spawn() behavior and whether Sonnet requests are subject to any additional auditing or retention, and (3) note a small metadata inconsistency — the _meta.json ownerId differs from the registry ownerId — if provenance matters, ask the publisher to clarify. If you handle sensitive inputs, consider disabling autonomous invocation or restricting the skill to non-sensitive contexts and testing with innocuous prompts first.Like a lobster shell, security has layers — review code before you run it.
latestvk97b9bnq0tjbr6xdm9n31w6g6h825m8f
License
MIT-0
Free to use, modify, and redistribute. No attribution required.
Runtime requirements
🧠 Clawdis