harness-sync

Merge metrics, observations, instincts, and failure logs across devices via claude-config.

Updated Mar 3, 2026
One-click install
npx skills add https://github.com/sukeun1997/claude-config --skill harness-sync
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: harness-sync
Source: https://github.com/sukeun1997/claude-config/tree/main/skills/harness-sync
Command: npx skills add https://github.com/sukeun1997/claude-config --skill harness-sync

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Harness data is spread across multiple environments (memory, homunculus, and topics), and this skill merges metrics, observations, instincts, and failure logs through the claude-config repository to maintain a unified dataset across devices.

Core Features & Use Cases

  • Push: Upload the current machine's memory/metrics, observations, instincts, and failure-log to sync-data in the repo.
  • Pull: Retrieve remote data from the repo and merge it into the local environment.
  • Status: Display the current synchronization state and remaining actions.
  • Data Mapping: Defines how sensors and logs (sessions.jsonl, agent usage, observations, instincts, failure-log, and skill-usage) are merged.
  • Merge Semantics: Uses JSONL dedup for metrics/observations and newer-mtime-wins for instincts, with conflict-handling rules described in the procedure.

Quick Start

Run harness-sync to push local harness data to the sync-data directory, or pull and merge updates from the repository.

Frequently Asked Questions about harness-sync

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

FAQPage Schema
How do I sync harness data across multiple devices using a git repository?

You can sync harness data by pushing local metrics, observations, instincts, and failure logs to a claude-config repository, then pulling and merging the remote data on other devices to maintain a unified dataset.

How does merging JSONL streams handle conflicts between observations and instincts?

Merging JSONL streams uses deduplication for metrics and observations, while instincts are resolved using a newer-mtime-wins time-based rule to ensure the most recent modification persists across sync operations.

What is the best way to track synchronization state for local and remote memory topics?

Running a status workflow displays the current synchronization state for memory, homunculus, and topics, showing remaining sync actions needed to align local harness data with the remote repository.

Can I use this skill to merge failure logs and agent usage data from different environments?

Yes, the skill defines data mapping for merging failure logs, agent usage, observations, and instincts from different environments into consistent, deduplicated JSONL streams within the claude-config repository.

What data formats are supported for pushing and pulling harness metrics in claude-config?

The skill processes JSONL streams for sessions, agent usage, observations, instincts, and failure logs, using deduplication and time-based resolution to merge these formats into a unified dataset.

Do I need any external dependencies to sync observations and instincts across devices?

No external dependencies are required; the skill operates independently to coordinate push, pull, and status workflows for syncing observations and instincts through the claude-config repository.