Langfuse 编码流程追踪

SkillMonitoring & ops

Provides Langfuse tracing for coding workflows. When coding skills such as logic-layer-method-impl run, it automatically reports start/end events for each major step (pattern recognition, code generation, compilation, launch, unit testing, code fixing) to a self-hosted Langfuse instance, enabling en

Use Langfuse 编码流程追踪 in Claude, ChatGPT or Ahel Desktop

Free. Sign in, add Langfuse 编码流程追踪 and connect your AI. About a minute.

Also: Claude Code · Cursor · Codex

Then ask your AI: use the Langfuse 编码流程追踪 skill

Details

Instructions available. Your AI can read the instructions. Execution depends on the setup they require.

Add ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.

Langfuse 编码流程追踪Start free

What this skill tells your AI

The instructions your AI receives, as published by hashgraph-online/awesome-codex-plugins in plugins/Colin4k1024/tsp/skills/langfuse-coding-trace/SKILL.md and read by ahel’s review.

为编码工作流每个关键步骤异步上报 Trace/Span 到 Langfuse,不阻塞主流程。

前置条件

需设置以下环境变量(未设置时静默跳过追踪):

LANGFUSE_PUBLIC_KEY   # Langfuse 公钥
LANGFUSE_SECRET_KEY   # Langfuse 私钥
LANGFUSE_HOST         # 自部署地址,如 http://192.168.1.100:3000

脚本位置

<skills_dir>/langfuse-coding-trace/scripts/trace.py

调用前先确认 LANGFUSE_PUBLIC_KEY 是否存在,不存在则跳过所有追踪调用。


追踪点规范

第一步:开始编码会话 → 创建 Trace(同步,获取 trace_id)

TRACE_ID=$(python "<skills_dir>/langfuse-coding-trace/scripts/trace.py" trace-start \
  --name "logic-layer-method-impl" \
  --input '{"method": "目标方法名", "mode": "首次编码/需求变更"}')
echo "Trace started: $TRACE_ID"

中间步骤:每个大步骤 → 创建 Span(异步,后台执行)

# 开始 Span(同步获取 span_id,耗时极短)
SPAN_ID=$(python "<skills_dir>/langfuse-coding-trace/scripts/trace.py" span-start \
  --trace-id "$TRACE_ID" \
  --name "maven-qa")

# 执行实际步骤 ...(编译/运行/测试等)

# 结束 Span(异步,后台执行)
python "<skills_dir>/langfuse-coding-trace/scripts/trace.py" span-end \
  --span-id "$SPAN_ID" \
  --output '{"exit_code": 0, "result": "通过"}' \
  --level DEFAULT &

# 若失败,用 ERROR 级别
python "<skills_dir>/langfuse-coding-trace/scripts/trace.py" span-end \
  --span-id "$SPAN_ID" \
  --output '{"error": "错误摘要"}' \
  --level ERROR &

最后一步:结束 Trace(异步)

python "<skills_dir>/langfuse-coding-trace/scripts/trace.py" trace-end \
  --trace-id "$TRACE_ID" \
  --output '{"status": "success", "loops": 1, "compile": "pass", "run": "pass", "test": "95%"}' &

标准追踪点列表

步骤Span 名称input 字段output 字段失败 level
识别编码模式identify-mode{}{"mode": "首次/变更"}WARNING
代码生成code-generation{"files": [...]}{"files_modified": N}ERROR
QA 质量闸口maven-qa{}{"compile": "pass", "pass_rate": "XX%", "criticalIssues": 0, "run": "pass"}ERROR
代码修复code-fix{"error": "..."}{"files_fixed": [...]}WARNING

完整调用示例(配合 logic-layer-method-impl)

# 检查环境变量
if [ -z "$LANGFUSE_PUBLIC_KEY" ]; then
  echo "Langfuse 未配置,跳过追踪"
  LANGFUSE_ENABLED=false
else
  LANGFUSE_ENABLED=true
fi

SCRIPT="<skills_dir>/langfuse-coding-trace/scripts/trace.py"

# 1. 开始 Trace
[ "$LANGFUSE_ENABLED" = true ] && \
  TRACE_ID=$(python "$SCRIPT" trace-start --name "logic-layer-method-impl" --input '{"method":"xxx"}')

# 2. 识别模式 Span
[ "$LANGFUSE_ENABLED" = true ] && \
  SPAN_MODE=$(python "$SCRIPT" span-start --trace-id "$TRACE_ID" --name "identify-mode")
# ... 执行识别逻辑 ...
[ "$LANGFUSE_ENABLED" = true ] && \
  python "$SCRIPT" span-end --span-id "$SPAN_MODE" --output '{"mode":"首次编码"}' &

# 3. 代码生成 Span(同上模式)

# 4. QA 验证 Span(编译 + 单测 + 静态分析)
[ "$LANGFUSE_ENABLED" = true ] && \
  SPAN_QA=$(python "$SCRIPT" span-start --trace-id "$TRACE_ID" --name "maven-qa")
# ... 调用 maven-qa skill ...
[ "$LANGFUSE_ENABLED" = true ] && \
  python "$SCRIPT" span-end --span-id "$SPAN_QA" --output '{"compile":"pass","pass_rate":"98%","criticalIssues":0,"run":"pass"}' &

# 5. 结束 Trace
[ "$LANGFUSE_ENABLED" = true ] && \
  python "$SCRIPT" trace-end --trace-id "$TRACE_ID" --output '{"status":"success"}' &

Overview

为编码工作流的每个关键步骤(识别模式、代码生成、编译、启动、单元测试、代码修复)异步上报 Trace/Span 到自部署 Langfuse,实现全流程可观测。环境变量未配置时静默跳过,不阻塞主流程。

本 skill 不执行编码逻辑,只做可观测性上报,配合其他编码类 skill 联合使用。

Signals

GitHub stars
1k
Forks
316
Last commit
Oct 2026
Advanced
Item type
skill
Key
langfuse-coding-trace
Source
github.com/hashgraph-online/awesome-codex-plugins