Agent Skills
› opensquilla/opensquilla
› skill-creator-smoke-test
skill-creator-smoke-test
GitHub内部工具，用于在元技能创建流程中执行G3/G4冒烟测试。通过正负样本模拟验证候选SKILL.md的有效性，支持LLM生成或确定性回退，确保技能质量。
Trigger Scenarios
作为meta-skill-creator DAG步骤被调用
需要验证新创建的元技能是否通过基础质量门禁
Install
npx skills add opensquilla/opensquilla --skill skill-creator-smoke-test -g -y
SKILL.md
Frontmatter
{
"name": "skill-creator-smoke-test",
"provenance": {
"origin": "opensquilla-original",
"license": "Apache-2.0"
},
"description": "Internal tool (not user-invocable). Called by meta-skill-creator as a DAG step (kind: agent) to run G3 (positive smoke) and G4 (negative smoke) gates against a candidate meta-skill SKILL.md. Cross-vendor: fixture-generation LLM != classifier LLM. Returns JSON.",
"description_zh": "内部工具（不可由用户直接调用）。由meta-skill-creator作为DAG步骤（kind: agent）调用，针对候选元技能SKILL.md运行G3（正向冒烟）和G4（负向冒烟）闸门。跨厂商：fixture生成LLM≠分类器LLM。返回JSON。",
"user-invocable": false,
"disable-model-invocation": true
}
Skill Creator Smoke Test
When invoked as a kind: agent, skill: skill-creator-smoke-test step, this sub-agent:
- Receives
skill_md,fixture_gen_model,classifier_modelfrom the parent step'swith: - Calls
simulate_meta_resolutionwith a positive fixture (LLM-generated usingfixture_gen_model) - Calls
simulate_meta_resolutionwith a negative fixture (LLM-generated, cross-domain) - Returns
{"G3": {...}, "G4": {...}}JSON
If OPENROUTER_API_KEY is absent OR either fixture_gen_model/classifier_model is unconfigured, the sub-agent falls back to the deterministic fixture generator and pins classifier_model="stub". In both cases the smoke step still emits G3/G4 records, but with a degraded: true flag.
Fallback
If the sub-agent fails entirely, output {"G3": {"passed": false, "reason": "smoke unavailable"}, "G4": {"passed": false, "reason": "smoke unavailable"}}.
Version History
- f662be3 Current 2026-07-31 11:49
- 7f72a32 2026-07-05 18:39
