dojo.md
University for AI agents. 92 courses, 4400+ scenarios, any model via OpenRouter. Auto-training loops generate per-model SKILL.md documents. Works with Claude Code, OpenClaw, Cursor, Windsurf. No fine-tuning req...
End-to-end trace debugging from trace_id. Fetch Jaeger trace and Elasticsearch logs, analyze possible bugs (optionally with local repository context), and generate a fixed-structure Markdown report for CI or tickets.
Follow the repository instructions for this skill and your agent. Review the files and dependencies before adding it to your workflow.
SkillHub CLI
skillhub install gakkiismywife/trace-debugerInspect the source and compatibility requirements, then follow its installation steps.
Give your agent the source link and ask it to inspect instructions and dependencies before explaining the installation steps.
Use this skill to generate a self-contained Markdown trace debug report.
trace_id (required)jaeger_url (optional, default http://127.0.0.1:16686)es_url (optional, default http://127.0.0.1:9200)repo_path (optional, absolute path, default /Users/noodles/Desktop/code/go-components/examples/tracer)output_path (optional, default ./trace_debug_report_{trace_id}.md)es_index (optional, default filebeat-tracer-*)es_size (optional, default 2000)python3 skills/trace_debuger/scripts/trace_debuger.py \
--trace-id <TRACE_ID> \
[--jaeger-url http://127.0.0.1:16686] \
[--es-url http://127.0.0.1:9200] \
[--repo-path /Users/noodles/Desktop/code/go-components/examples/tracer] \
[--output-path ./trace_debug_report_<TRACE_ID>.md]
output_path
【markdown报告文件】 is a placeholder and MUST be replaced with the real uploaded Markdown filename (example: trace_debug_report_<trace_id>.md).
<真实markdown报告文件名>
trace_id: xxxx
status: xxx
jaeger_url: xxx
es_url: xxx
代码仓库路径:仓库路径
关键结论摘要:xxxx
trace_id: <trace_id>
status: SUCCESS/FAIL
jaeger_url: <jaeger_url>
es_url: <es_url>
代码仓库路径:<repo_path|N/A>
关键结论摘要:<summary>
codex exec equivalent to TUI paste workflow) with this prompt:
这是我的日志,请根据日志结合代码帮我排查分析bug,输出bug原因及解决方案,必须保持固定的格式。
These signals come from indexed source analysis and may lag behind changes. Check the current source files.
# Trace Debuger
Use this skill to generate a self-contained Markdown trace debug report.
## Inputs
- `trace_id` (required)
- `jaeger_url` (optional, default `http://127.0.0.1:16686`)
- `es_url` (optional, default `http://127.0.0.1:9200`)
- `repo_path` (optional, absolute path, default `/Users/noodles/Desktop/code/go-components/examples/tracer`)
- `output_path` (optional, default `./trace_debug_report_{trace_id}.md`)
- `es_index` (optional, default `filebeat-tracer-*`)
- `es_size` (optional, default `2000`)
## Run
```bash
python3 skills/trace_debuger/scripts/trace_debuger.py \
--trace-id <TRACE_ID> \
[--jaeger-url http://127.0.0.1:16686] \
[--es-url http://127.0.0.1:9200] \
[--repo-path /Users/noodles/Desktop/code/go-components/examples/tracer] \
[--output-path ./trace_debug_report_<TRACE_ID>.md]
```
## Output
- Writes Markdown report to `output_path`
- **MUST send the generated Markdown report to the user as a file attachment via the chat window in the same session before finishing the task**
- **MUST send the report as ONE chat message only: attach the Markdown file and put the strict summary block in the same message caption/body.**
- **`【markdown报告文件】` is a placeholder and MUST be replaced with the real uploaded Markdown filename (example: `trace_debug_report_<trace_id>.md`).**
- **Use exactly this format in caption/body:**
```text
<真实markdown报告文件名>
trace_id: xxxx
status: xxx
jaeger_url: xxx
es_url: xxx
代码仓库路径:仓库路径
关键结论摘要:xxxx
```
- Prints fixed summary lines to stdout:
```text
trace_id: <trace_id>
status: SUCCESS/FAIL
jaeger_url: <jaeger_url>
es_url: <es_url>
代码仓库路径:<repo_path|N/A>
关键结论摘要:<summary>
```
## Notes
- Keep logs sorted by timestamp ascending.
- After fetching ES logs, run Codex in repository root (automated via `codex exec` equivalent to TUI paste workflow) with this prompt:
- `这是我的日志,请根据日志结合代码帮我排查分析bug,输出bug原因及解决方案,必须保持固定的格式。`
- If repositor...University for AI agents. 92 courses, 4400+ scenarios, any model via OpenRouter. Auto-training loops generate per-model SKILL.md documents. Works with Claude Code, OpenClaw, Cursor, Windsurf. No fine-tuning req...
macOS WPS Office workflow helper skill for safer document preparation, conversion, export, and compatibility guidance
A股分析技能
A2A4B2B Skill for OpenClaw