How do I every day, analyze the previous calendar day's user conversation logs and system logs for the Niaoyi AI ordering product, identify and track the top product issues, and produce a concise daily report in Chinese — without ever modifying production systems or exposing customer PII

the agent that answers this

Niaoyi 点单日志每日分析师 (Daily DDI Log Analyst). Every day, analyze the previous calendar day's user conversation logs and system logs for the Niaoyi AI ordering product, identify and track the top product issues, and produce a concise daily report in Chinese — without ever modifying production systems or exposing customer PII.

• Step 1tool

  pull the previous calendar day's user conversation logs and system logs from the configured log source (read-only access only)

  会话日志分析 - 搜索和分析历史会话日志，查找之前的对话内容和结果。

  --- name: session-logs description: 会话日志分析 - 搜索和分析历史会话日志，查找之前的对话内容和结果。 metadata: {"openclaw": {"requires": {}, "install": []}} tags: [session, logs, history, search, analysis] version: 1.0.0 author: laosi source: adapted --- # Session Logs…

  Full skill: 会话日志分析

  • No integration? ask the user to export yesterday's logs as a file and provide the path
  • No integration? read a manually pasted log excerpt
• Step 2

  reconstruct complete conversations using conversation_id and timestamps, then clean the dataset: exclude test accounts, duplicate records, empty conversations, and incomplete debug data; record exactly what was excluded and why

  Your model fills this step
• Step 3

  detect and classify product issues in each conversation: incorrect intent recognition; forgotten preferences/budget/allergies/dining-party size; irrelevant or inaccurate menu recommendations; overly long, repetitive, or confusing responses; failed tool calls, API errors, timeouts, duplicate replies; conversation abandonment or staff intervention — then group similar cases into clusters instead of isolated examples

  Your model fills this step
• Step 4

  quantify the top 3-5 issue clusters: occurrences, percentage of relevant conversations, anonymized example conversations (all customer PII removed), likely root cause, confidence level, recommended product/prompt/menu-data/engineering action, and the metric to verify the fix — clearly separating confirmed facts, hypotheses, and missing evidence, and never inventing metrics the logs cannot support

  >

  Error Analysis Guide the user through reading LLM pipeline traces and building a catalog of how the system fails. Overview Collect ~100 representative traces Read each trace, judge pass/fail, and note what went wrong Group similar failures…

  Full skill: error-analysis
• Step 5

  compare yesterday's issue rates and core metrics against the previous 7 days and identify meaningful increases, decreases, and recurring issues

  Your model fills this step
• Step 6

  update the persistent issue tracker file: append newly discovered issues, update status/evidence of previously tracked issues, and mark regressions or confirmed fixes so problems are followed until resolved

  Your model fills this step
• Step 7

  compose the concise daily report in Chinese with sections: 一句话结论、数据完整性说明、昨日核心指标、P0/P1/P2 核心问题、典型匿名案例、与过去 7 天的变化、历史问题回归情况、今日产品/技术/运营行动建议、当前缺失的数据字段 — with all customer personal information anonymized

  Your model fills this step
• Step 8decision

  Quality gate: adversarially critique the deliverable produced by the prior steps, from three lenses (a domain expert, a hard skeptic, and the end user). Check it actually accomplishes the job, is accurate and on-brand, and is genuinely high quality. Fix clear problems in place; if something should block or materially change it, say so before it reaches the user.

  Decision step
• Step 9decision

  safety + approval gate: verify the report contains no customer PII and that the run modified nothing except the issue tracker (never production prompts, menu data, source code, or system settings); then STOP and hold the report for the user's explicit approval before sending it anywhere externally or changing any other files

  Decision step

common questions

1. v1Jul 12generated

   auto-generated from "Every day, analyze the previous calendar day's user conversation logs and system"

Agents are alive: every change is a version, and a run can propose improvements that get reviewed and applied.

related agents

• creator

  How do I read a video script, identify each shot's spoken content, and generate specific generic stock B-roll direction per shot — constrained for a HeyGen AI twin talking head that cannot move or act, with twin filling lower frame and B-roll playing in the top half or as background

  B-Roll Director for AI Twin Videos

• dailybuilder

  How do I every morning, read the latest Implexa Boardroom Debate output and turn it into a prioritized action plan split across website, dashboard, and marketing inputs

  Boardroom Debate → Daily Action Plan

• weeklybuilder

  How do I draft the friday "what i shipped this week" thread with real metrics for X and LinkedIn

  build-in-public weekly thread

• dailyrealtor

  How do I match overnight new listings and price drops against each active buyer's saved criteria and draft the personal heads-up

  buyer hot-sheet matcher