AgentSkillsCN

log-triage

无需将所有日志文件一次性加载至上下文，即可通过提取与归类错误信息、生成精简的调试摘要，并（可选）将原始日志分块上传至 LLM，从而高效地对大型日志目录进行分类与处理。

SKILL.md
--- frontmatter
name: log-triage
description: Triage large log directories without loading everything into context by extracting and grouping errors, emitting a compact debug brief, and (optionally) chunking raw logs for LLM upload.

log-triage

Purpose

Work with very large log files by producing grepable, human-readable artifacts:

  • •an index of log files + sizes
  • •a compact error report (top error signatures, example lines, nearby context)
  • •optional raw chunks of the original logs for LLM upload

This is meant to avoid pasting entire logs into a prompt.

Assumptions

  • •Logs are plaintext. Mixed encodings are tolerated (invalid bytes replaced).
  • •“Errors” are approximated using regex heuristics; this is not a full structured log parser.
  • •The script should be runnable without network access.

Workflow

  1. •Choose an input directory (example: backend/logs).
  2. •Run triage to generate:
    • •triage/index.json
    • •triage/errors.txt
    • •triage/errors.json
    • •triage/samples/<hash>.txt (context windows)
  3. •If needed, chunk specific log files for LLM upload (using the existing log-to-llm-chunker skill).

Quick start

  • •

    Triage everything in backend/logs into backend/logs/triage:

    • •python .github/skills/log-triage/scripts/log_triage.py --input backend/logs --output backend/logs/triage
  • •

    If you want to chunk a log for LLM upload, use:

    • •python .github/skills/log-to-llm-chunker/scripts/chunk_log_for_llm.py --input backend/logs/app.log --output backend/logs/LLMreadablelogs
  • •

    Limit to specific filename patterns:

    • •python .github/skills/log-triage/scripts/log_triage.py --input backend/logs --output backend/logs/triage --include "*.log*"

Output

  • •index.json: file inventory + sizes.
  • •errors.json: grouped error signatures with counts, first/last seen file, example line.
  • •errors.txt: human scan format version of errors.json.
  • •samples/: small context windows around representative error lines.
  • •Raw log chunking is handled by log-to-llm-chunker.

Debug usage

  • •Start from errors.txt. Pick the highest-frequency signature and jump to displayed sample files.
  • •For connection-style errors, verify the downstream service is reachable and that host/port/version match.
  • •If errors are repetitive, add one-shot abort/backoff to the caller.