fill-evidence

Parse Model A/B logs into structured execution evidence files with runtime summaries.

Updated Aug 27, 2026

Search Tags:#logs #runtime #evidence #checkpoint #data-collection #test-results #build-output

Ready to run

npx skills add https://github.com/MinhOmega/marlin-skill --skill fill-evidence

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: fill-evidence
Source: https://github.com/MinhOmega/marlin-skill/tree/main/.claude/skills/fill-evidence
Command: npx skills add https://github.com/MinhOmega/marlin-skill --skill fill-evidence

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Turning raw Model A/B logs into structured execution evidence files, saving time during checkpoint reviews.

Core Features & Use Cases

Extracts test results, build output, runtime, and changed files from turn logs.
Generates dedicated evidence files for Model A and Model B to streamline evaluation.
Supports quick post-run validation and reporting for multi-turn experiments.

Quick Start

After saving logs with step_4_save_logs.sh, run /fill-evidence <workspace_path> <turn_number> to generate evidence files.

Frequently Asked Questions about fill-evidence

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I generate execution evidence from Model A and B logs for checkpoint reviews?▼

To generate execution evidence from logs, parse the Model A/B text files to extract test results, build output, runtimes, and changed files into structured markdown summaries. This produces dedicated evidence files that streamline checkpoint assessment.

How do I extract test results and build output from turn-level experiment logs?▼

Extract test results and build output by applying the evidence generation skill to a workspace path and turn number. It reads the logs_a.txt and logs_b.txt files in the turn directory to produce structured execution evidence markdown files.

What is the best way to summarize runtime and changed files from raw logs for post-run validation?▼

The best way to summarize runtime and changed files is to parse the raw logs into structured evidence files. This process extracts execution metrics and file change data, generating concise summaries for quick post-run validation and reporting.

Do I need to save logs with step_4_save_logs.sh before generating execution evidence files?▼

Yes, you need to save logs using step_4_save_logs.sh before generating evidence files. The evidence generation process requires logs_a.txt and logs_b.txt to be present in the turn directory to extract the test results and runtime data.

Can I use this evidence generation process for multi-turn experiments without manual log parsing?▼

Yes, you can automate evidence generation for multi-turn experiments by running the skill with the workspace path and turn number. It automatically parses the existing logs to produce structured execution evidence files without manual extraction.

What limitations exist when parsing logs to produce structured execution evidence for Model A and B?▼

A key limitation is that the logs must be present in the turn directory as logs_a.txt and logs_b.txt. Evidence generation only works for turn-level reviews where these specific raw log files already exist from prior runs.