logs-analysis

name	logs-analysis
description	Analyze iteration logs in ./logs for errors, schedule gaps, and problems. Checks if cron jobs (run.sh, review-prs, add-backlog, learnable-pattern-search, check-signal) are running correctly. Triggers on: analyze logs, check logs, logs analysis, log report.

Logs Analysis

Analyze iteration log files in ./logs/ (top-level only, never look in processed/) and cron logs in ./logs/*-cron.log to identify problems and assess schedule health.

Determine the bot directory: This SKILL.md file lives inside the bot's .claude/skills/logs-analysis/ directory. Use this file's path to derive the absolute path to the bot directory (referred to as $BOT_DIR below).

---

Step 1: Gather Log Inventory

List all .log files in $BOT_DIR/logs/ (top-level only, exclude processed/ subdirectory):

find $BOT_DIR/logs -maxdepth 1 -name '*.log' -type f | sort

Also list the cron logs:

ls -la $BOT_DIR/logs/*-cron.log 2>/dev/null

If there are no log files in the top-level logs directory, report "No unprocessed logs found" and stop.

---

Step 2: Analyze Logs Using Subagents

To handle large numbers of logs efficiently, batch log files into groups of up to 10 and spawn one subagent per batch. Run all subagents in parallel.

For each batch, launch an Agent (subagent_type: "general-purpose") with a prompt like:

Analyze the following iteration log files for problems. For each log file:

1. Read the file
2. Check for errors: "command not found", API errors (500, connection refused, socket errors), permission denied, fatal errors, crash/panic, "integer expression expected", unexpected exits
3. Check what story was worked on and whether it completed successfully or failed
4. Note the start/end timestamps if available
5. Note the file size (very small files like <100 bytes usually indicate a problem like "command not found")

Return a structured summary for each file:

• Filename
• Story ID worked on (or "unknown")
• Status: SUCCESS / FAILURE / EMPTY (for tiny files)
• Errors found (list each error with context)
• Duration if determinable

Files to analyze: [list of file paths]

Also spawn a separate subagent for cron log analysis:

Analyze the cron log files for problems. For each file:

1. Read the tail (last 200 lines) of the file
2. Check for errors, failures, and the last successful run timestamp
3. Check for schedule gaps (compare timestamps of runs against expected cron schedule)

Expected cron schedule:

• run-cron.log: runs at 04:00 and 13:00 daily (EST, which is UTC-5)
• review-prs-cron.log: runs at 05:00, 08:00, 10:00, 12:00, 14:00, 16:00, 18:00, 20:00 daily (EST)
• add-backlog-cron.log: runs at 03:00 and 11:00 daily (EST)
• learnable-pattern-search-cron.log: runs at 06:00 daily (EST)
• check-signal-cron.log: runs every hour at :05 (EST)

Report:

• Last run timestamp for each cron job
• Whether each schedule appears to be running on time
• Any errors or anomalies in the cron logs
• Whether there are gaps (missed runs)

Cron log files: [list of cron log paths]

---

Step 3: Compile Report

After all subagents return, compile a unified report with these sections:

Schedule Health

• For each cron job: last run time, whether it's on schedule, any missed runs
• Flag any job that hasn't run in over 24 hours as a problem

Iteration Log Summary

• Total logs analyzed
• Success/failure/empty counts
• Table of each log with: filename, story, status, key errors

Problems Found

• Group all errors by type (command not found, API errors, permission errors, etc.)
• Highlight recurring patterns
• Flag critical issues that need immediate attention

Recommendations

• Actionable suggestions based on the problems found

---

Step 4: Ask to Archive

After presenting the report, ask the user:

Would you like to move the analyzed log files to logs/processed/?

If the user agrees, move all analyzed log files (from the top-level logs/ directory only) into logs/processed/:

mv $BOT_DIR/logs/*.log $BOT_DIR/logs/processed/

---

Important Rules

• NEVER read or analyze files in logs/processed/ — those have already been reviewed
• NEVER delete log files — only move them to processed/
• Keep the report concise but include enough detail to diagnose issues
• If a log file is very large (>1MB), instruct subagents to focus on the first 100 and last 100 lines
• For stream-json format logs, errors often appear in "type":"error" or "type":"result" entries — grep for those patterns