Quality Boost - 大模型回答质量提升器
v1.0.0通过9条严格规则系统评估并量化大模型回答质量，防止幻觉、跑偏、冗长和遗漏，确保准确、简洁、合规的输出。
⭐ 0· 61·0 current·0 all-time
byjjx@jx-76
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
Name/description (improving LLM answer quality) matches the provided materials: 9 rules, templates, quick-apply snippets, and a system-prompt file. The skill requires no binaries, env vars, or external services — all requested resources are consistent with a prompt/policy helper.
Instruction Scope
SKILL.md and supporting files are a set of prompts and templates to be applied to conversations. They remain within the stated scope, but include an explicit 'system injection' template and instructions for adding a system prompt. That changes agent behavior globally if applied; it's expected for this purpose but broadens the skill's runtime impact (affects all future responses when injected).
Install Mechanism
No install spec or external downloads — instruction-only. No code files to execute, and the regex scanner had nothing to analyze, so no installation risk was found.
Credentials
No environment variables, credentials, or unrelated config paths are requested. The only file path referenced is ~/.openclaw/config.json and a skill file under ~/.openclaw/skills — proportional to the stated goal of optionally persisting behavior.
Persistence & Privilege
The skill does not force 'always: true', but it provides instructions to modify the user's OpenClaw config (systemPrompt prepend and an autoApply example). Writing to the agent's own config to persist a system prompt is coherent for this skill, but it is a persistent change that will affect future agent behavior until the user reverts it.
Assessment
This skill is an instruction-only prompt/template pack and appears to do what it says. Before applying it globally: (1) review the system-inject content — injecting it will prepend a system prompt that affects all agent responses; (2) back up ~/.openclaw/config.json before making changes so you can revert; (3) prefer manual invocation or session-level use first (use the quick-apply templates) to confirm the behavior meets your needs before enabling autoApply/system injection; (4) never paste the templates into external services unless you intend those services to adopt the rules. No credentials or external downloads are required.Like a lobster shell, security has layers — review code before you run it.
latestvk97ae5746s10f72qak90c2emgs8426ej
License
MIT-0
Free to use, modify, and redistribute. No attribution required.