harness-curator

Cluster structured harness friction reports into labeled GitHub issues.

20|4|Updated May 14, 2026
One-click install
npx skills add https://github.com/crewrig/crewrig --skill harness-curator-crewrig
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: harness-curator
Source: https://github.com/crewrig/crewrig/tree/main/artifacts/library/skills/harness-curator
Command: npx skills add https://github.com/crewrig/crewrig --skill harness-curator-crewrig

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires bash, jq, gh, glab, tea, mempalace, and includes scripts (resource) and assets (resource) components.

What problem does it solve?

This Skill closes the feedback loop for AI-assisted work by collecting friction reports, grouping related failures, and surfacing them as clear GitHub issues instead of leaving them buried in local logs.

Core Features & Use Cases

  • Friction clustering: Reads structured friction payloads from the harness wing and groups them by subcategory or room.
  • Issue generation: Composes descriptive issue titles, markdown bodies, and labels for each qualifying cluster.
  • Routing and deduplication: Selects the correct target repository from canonical provenance, skips already-correlated drawers, and avoids reopening existing clusters when dedup mode is enabled.
  • Deep sweep review: Scans transcript drawers with heuristics to produce a triage-friendly Markdown review before promotion into real friction reports.
  • Automation support: Includes local scheduling, label bootstrapping, and dry-run modes for safe validation before apply.

Quick Start

Run the harness-curator skill to analyze the friction wing and either preview or open clustered issues.

Frequently Asked Questions about harness-curator

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

FAQPage Schema
How do I automatically create GitHub issues from harness friction logs?

Clustering structured friction payloads from the harness wing groups related failures by subcategory or room, surfacing them as clear GitHub issues with descriptive titles, markdown bodies, and labels instead of leaving them buried in local logs.

What is friction clustering in AI workflow triage?

Friction clustering reads structured payloads from the harness wing and groups related failures by subcategory or room, composing descriptive issue titles, markdown bodies, and labels for each qualifying cluster to enable actionable AI workflow triage.

How do I deduplicate recurring GitHub issues from automated triage runs?

Routing and deduplication selects the correct target repository from canonical provenance, skips already-correlated drawers, and avoids reopening existing clusters when dedup mode is enabled during recurring curator runs and deep transcript sweeps.

Can I preview clustered GitHub issues before opening them on a repository?

Yes, dry-run modes generate triage-friendly markdown reviews of transcript drawers using heuristics, allowing you to safely validate and preview clustered GitHub issues before promoting them into real friction reports.

Do I need the mempalace Python package to generate GitHub issues from friction reports?

Yes, mempalace is required to read local wing data and open labeled issues safely, alongside bash, jq, and a forge CLI like gh, glab, or tea for issue generation across supported automation workflows.

What are the limitations of using bash and jq for automated friction issue generation?

Automated friction issue generation requires a local wing data source, a configured forge CLI like gh or glab for repository access, and Python with the mempalace package installed to safely handle structured payloads and label bootstrapping.