src/dynamic_os/skills/builtins/aggregate_results/SKILL.md
# aggregate_results 聚合多轮实验结果,生成跨迭代对比分析。 ## 功能 - 从所有 ExperimentResults 和 ExperimentIteration 产物中提取指标历史 - 构建迭代对比表(每轮指标 + 成功/失败状态) - 计算各指标的 min/max/avg/best 统计值 - 提取最佳配置和经验教训 - 调用 LLM 生成结构化分析文本 ## 适用场景 当实验循环结束(ExperimentIteration.should_continue=false)且存在多轮实验数据时, 由 analyst 角色调用此技能生成统一的分析报告,供 draft_report 使用。
npx skillsauth add szaaaaaa/researchagent src/dynamic_os/skills/builtins/aggregate_resultsInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
聚合多轮实验结果,生成跨迭代对比分析。
当实验循环结束(ExperimentIteration.should_continue=false)且存在多轮实验数据时, 由 analyst 角色调用此技能生成统一的分析报告,供 draft_report 使用。
tools
Clarify the user's intent into one of three tiers: a confidently inferred goal, a default-filled goal with explicit assumptions, or a structured reask with option-style questions. The skill uses `ctx.tools.llm_chat()` with a structured JSON schema and emits either `ClarifiedIntent` or `ClarificationRequest`. When previous-round `ClarificationResponse` artifacts are present in `ctx.input_artifacts`, they are folded into the prompt so the LLM can incorporate prior answers.
data-ai
# extract_keywords_v0 (broken baseline) Demo asset for the `reflect_on_failure → optimize_skill` chain. See `run.py` for the deliberate bug.
testing
Use this skill to search papers.
testing
This fixture skill attempts a network call and should be blocked by its own permissions.