Does this skill overwrite existing logs?

No, it strictly appends to the log file to ensure a permanent, chronological record of every run, including discarded, kept, or failed attempts.

Can I track different types of metrics?

Yes, the skill supports a primary optimization metric and multiple secondary metric key-value pairs, allowing for comprehensive performance analysis of every run.

What are segments in the context of session persistence?

Segments group experiments under specific baselines or configurations. They increment automatically when optimization targets change, allowing you to compare results within distinct phases of research.

What happens if Claude Code crashes during an experiment?

The session-persistence skill reads the autoresearch.jsonl file upon restart to rebuild the session state, run counts, and baseline metrics, ensuring no experiment data is lost.

Where is the session data stored?

All data is appended to a local file named autoresearch.jsonl in a structured JSONL format, making it easy to parse, audit, or move between environments.

Session Persistence for Autoresearch

Name: Session Persistence for Autoresearch
Author: pbdeuchler

bypbdeuchler

•

분석 및 모니터링

Manages persistent logging and session recovery for automated code experiments and performance optimization loops.

This skill provides robust state management for the autoresearch plugin by maintaining a canonical record of experiments in a JSONL format. It enables Claude to survive context resets, discards, or crashes by automatically rebuilding the session state, including baselines, metrics, and segment history from the local log file. By handling configuration headers and detailed experiment records, it ensures that every optimization attempt is recorded and verifiable, allowing for complex multi-stage research workflows that persist across sessions.

주요 기능

01Automatic session recovery and state reconstruction from local files

02Detailed recording of primary metrics, secondary metrics, and confidence scores

03Durable JSONL logging that survives crashes and context resets

04Segment-based tracking for multi-stage optimization targets

05Real-time run summaries with delta comparisons against baselines

0611 GitHub stars

사용 사례

01Recovering optimization state after an unexpected LLM disconnect or context wipe

02Auditing automated code changes through chronological experiment logs

03Maintaining experiment history during long-running performance tuning sessions

주요 기능

01Automatic session recovery and state reconstruction from local files

02Detailed recording of primary metrics, secondary metrics, and confidence scores

03Durable JSONL logging that survives crashes and context resets

04Segment-based tracking for multi-stage optimization targets

05Real-time run summaries with delta comparisons against baselines

0611 GitHub stars

사용 사례

01Recovering optimization state after an unexpected LLM disconnect or context wipe

02Auditing automated code changes through chronological experiment logs

03Maintaining experiment history during long-running performance tuning sessions