Add Apache error log analysis tasks

[ScuttleBot]

Apache Error Log Analysis Tasks

Adds 5 new tasks for analyzing an Apache error log (assets/logs/apache_error.log):

1. task_log_apache_client_issues - Identify the most problematic client IPs by error count, including awstats scanners and IIS worm probes
2. task_log_apache_top_errors - Rank and categorize all error types from most to least frequent
3. task_log_apache_error_summary - Generate a comprehensive summary report covering server config issues, client errors, and security assessment
4. task_log_apache_critical - Identify critical security issues with severity classification (critical/high/medium/low)
5. task_log_apache_timeline - Create a day-by-day timeline identifying error spikes and the peak burst event

All tasks use the same Apache error log asset (1000 lines, Jun 9-16 2005) containing a rich mix of:

• Directory scanning and enumeration
• IIS worm probes (Nimda/Code Red variants)
• Awstats vulnerability scanning
• FrontPage extension probing
• Server configuration errors (mod_jk, JK connector)
• Buffer overflow attempts (URI too long)

Closes #209, Closes #210, Closes #211, Closes #212, Closes #213

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Solid addition of 5 well-structured log analysis tasks. The grading functions handle edge cases (missing files, JSON parse errors) gracefully, the expected values are well-documented, and the task prompts are clear and actionable.

Files Reviewed (6 files)

• tasks/manifest.yaml
• tasks/task_log_apache_client_issues.md
• tasks/task_log_apache_top_errors.md
• tasks/task_log_apache_error_summary.md
• tasks/task_log_apache_critical.md
• tasks/task_log_apache_timeline.md

---

_{Reviewed by claude-4.6-sonnet-20260217 · 112,449 tokens}

[ScuttleBot]

🧪 PR Test Started

Instance: 155.138.235.245 (Vultr vc2-2c-4gb, ATL)
Branch: tasks/log-apache

Models

#	Model
1	openrouter/anthropic/claude-opus-4.6
2	openrouter/openai/gpt-5.4
3	openrouter/google/gemini-2.5-pro

Tasks (5 new Apache log tasks)

• task_log_apache_client_issues
• task_log_apache_top_errors
• task_log_apache_error_summary
• task_log_apache_critical
• task_log_apache_timeline

ETA: ~30-45 minutes (3 models running in parallel)
Started: 2026-04-15 08:33 ET

[ScuttleBot]

🧪 PR Test Results — Apache Error Log Tasks

Instance: 155.138.235.245 (Vultr vc2-2c-4gb, ATL)
Branch: tasks/log-apache
Duration: ~24 min (all 3 models in parallel)

⚠️ Bug Found & Fixed During Test

The task definitions used path instead of dest in workspace_files, and source included the assets/ prefix which gets double-prepended by the benchmark runner. All 5 tasks failed with KeyError: 'dest' until fixed:

workspace_files:
-  - path: "apache_error.log"
-    source: "assets/logs/apache_error.log"
+  - dest: "apache_error.log"
+    source: "logs/apache_error.log"

This fix needs to be applied to all 5 task files before merging.

---

Overall Scores

Model	Score	Pct
openrouter/openai/gpt-5.4	4.8 / 5.0	96.0%
openrouter/google/gemini-2.5-pro	2.6 / 5.0	52.7%
openrouter/anthropic/claude-opus-4.6	1.9 / 5.0	39.0%

Per-Task Breakdown

Task	Opus 4	GPT-5.4	Gemini 2.5 Pro
task_log_apache_client_issues	50%	93%	100%
task_log_apache_top_errors	45%	92%	45%
task_log_apache_error_summary	0%	95%	0%
task_log_apache_critical	100%	100%	98%
task_log_apache_timeline	0%	100%	20%

Observations

task_log_apache_error_summary — Both Opus and Gemini scored 0% because output_created was 0. The agents may have written to a different filename or path than expected (error_summary.md). GPT-5.4 got 95% on it, so the task spec is workable but may need clearer output filename guidance.

task_log_apache_top_errors — Opus (45%) and Gemini (45%) both struggled. The automated grading for Gemini shows output_created: 0 suggesting a similar file path issue. May need to verify the expected output filename matches what's in the prompt.

task_log_apache_timeline — Opus 0%, Gemini 20%. Both had output_created: 0 (Opus) or failed on all sub-criteria except output creation (Gemini). The peak burst identification and daily breakdown requirements may be quite strict.

task_log_apache_critical — All 3 models scored 98-100%. Excellent task — well-calibrated difficulty.

task_log_apache_client_issues — GPT-5.4 (93%) and Gemini (100%) did well. Opus at 50% may have missed some IP ranking details.

Timing

Model	Duration
Opus 4	23m 34s
GPT-5.4	18m 21s
Gemini 2.5 Pro	21m 13s

Recommendations

1. Fix workspace_files format in all 5 tasks (see diff above) — blocker
2. Review task_log_apache_error_summary and task_log_apache_top_errors output filename expectations — 2 of 3 models wrote output but grader couldn't find it
3. task_log_apache_critical is well-calibrated (all models pass) ✅
4. task_log_apache_timeline may need tuning — strict criteria caused low scores across models