Get Logs

Step 1: Check tmux session exists

Run:

tmux has-session -t <session_name> 2>/dev/null && echo "EXISTS" || echo "NOT_FOUND"

If NOT_FOUND:

• List all sessions: tmux list-sessions
• Suggest closest session name
• Report clear error

Step 2: Capture logs from tmux

# Capture last N lines from buffer
tmux capture-pane -t <session_name> -p -S -<N>

If more than default buffer needed:

# Capture entire scrollback buffer
tmux capture-pane -t <session_name> -p -S -

If --grep specified:

tmux capture-pane -t <session_name> -p -S - | grep -E "<pattern>"

Step 3: Analyze status

Determine model status from logs:

RUNNING:

• Active cursor (new output appearing)
• Signs of processing: "Running...", "Building...", "Testing..."
• No completion indicators

COMPLETED:

• Completion indicators: "Done", "Finished", "Complete", exit code 0
• Test summary present: "X passed, Y failed"
• Shell prompt returned ($, %, >, #)

ERROR:

• Error messages: "Error:", "FAIL", "Exception", "Traceback"
• Non-zero exit code
• Process killed or timed out
• Segfault, OOM, or crash signals

IDLE:

• Session exists but no process running
• Shell is idle

Step 4: Extract key information

Auto-detect and extract:

Test Results:

• Pattern: X passed, X failed, X errors
• Pattern: PASS, FAIL, ERROR with test names
• Pattern: pytest, jest, go test, cargo test output
• Summary: total tests, passed, failed

Errors and Warnings:

• Lines containing: Error, error, ERROR, Exception, Warning, WARN
• Traceback blocks (Python)
• Compile errors
• Runtime errors

Runtime:

• Time from first to last line (if timestamps present)
• Or estimate from tmux session creation time

Files Changed:

• Pattern: git diff --stat, modified:, new file:
• List of files model changed

Step 5: Save logs (if --save)

tmux capture-pane -t <session_name> -p -S - > <save_path>

Add metadata header:

# Logs captured from tmux session: <session_name>
# Captured at: <timestamp>
# Lines: <count>
# Status: <RUNNING/DONE/ERROR>
---
<logs content>

Step 6: Compare (if --compare)

When comparing 2 sessions:

### Comparison: Model A vs Model B

| Criteria | Model A | Model B |
|----------|---------|---------|
| Status | COMPLETED | RUNNING |
| Runtime | 5m 23s | 3m 10s (running) |
| Tests passed | 25/25 | 18/20 (2 failed) |
| Errors | 0 | 2 |
| Files changed | 5 | 7 |

### Key Differences
- Model A completed first with 0 errors
- Model B has 2 test failures: [test names]
- Model B changed more files (7 vs 5)

Common Patterns

Quick check if model is running

/get-logs model_a

Continuous monitoring (check every few minutes)

/get-logs model_a --lines 20
# ... wait a few minutes ...
/get-logs model_a --lines 20

Get only test results

/get-logs model_a --grep "PASS\|FAIL\|passed\|failed\|error"

Save full logs for evidence

/get-logs model_a --save workspace/329_.../turn_1/logs_a.txt
/get-logs model_b --save workspace/329_.../turn_1/logs_b.txt

Quick compare 2 models

/get-logs model_a model_b --compare

Check specific errors

/get-logs model_b --grep "Traceback\|Exception\|Error"

Workflow Integration

In the checkpoint review flow:

1. Before collecting diffs: Check both models finished

   /get-logs model_a
   /get-logs model_b
   

2. Save logs as evidence: Logs serve as execution_evidence

   /get-logs model_a --save workspace/.../turn_N/logs_a.txt
   

3. Extract runtime: Get execution time for evaluation

   /get-logs model_a --grep "real\|user\|sys\|elapsed"
   

4. Debug model failures: Check why a model failed

   /get-logs model_b --grep "error\|fail" --lines 50
   

Notes

• Tmux buffer is limited (default 2000 lines). If model produces lots of output, old buffer is lost.
• To increase buffer: tmux set-option -g history-limit 50000
• For full logs, redirect output when running model: command 2>&1 | tee logs.txt
• This skill only READS logs, it does not interfere with running processes.