skills.rest
burningportra
burningportraCOMMUNITY
Explore all skills by this author →
GitHub

flywheel-refine-skills

Reviews and improves agent skills using session patterns and bead completion feedback.

3|1|Updated Apr 15, 2026
Search Tags:#code review#claude code#multi-agent#skill refinement#workflow improvement#background agents#agent mail
One-click install
npx skills add https://github.com/burningportra/agent-flywheel-plugin --skill flywheel-refine-skills-burningportra
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: flywheel-refine-skills
Source: https://github.com/burningportra/agent-flywheel-plugin/tree/main/skills/flywheel-refine-skills
Command: npx skills add https://github.com/burningportra/agent-flywheel-plugin --skill flywheel-refine-skills-burningportra

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Agent skills drift out of date as workflows evolve, and there is no systematic way to improve them based on real usage. This Skill analyzes session patterns, agent-mail history, and bead completion outcomes to propose concrete improvements to every skill in the skills/ directory. ## Core Features & Use Cases - Parallel skill analysis: Spawns one background analysis agent per skill via a coordinated Agent Mail team, each writing proposed changes to a docs file. - Evidence-driven refinement: Searches agent-mail history for skill feedback and planning patterns, and reads closed bead data from br list --json to ground improvements in actual outcomes. - Human-in-the-loop approval: Presents proposed changes per skill and asks which to apply before editing any SKILL.md file. - Use Case: After several flywheel sessions, run this Skill to have agents review all loaded skills against recent session feedback, then approve updates to the ones that underperformed. ## Quick Start Ask the agent to refine all skills based on recent session patterns and bead feedback, then approve the proposed changes you want applied.

Frequently Asked Questions about flywheel-refine-skills

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

FAQPage Schema
How do I improve agent skills based on session feedback?▼

Run the refine-skills workflow, which searches agent-mail history for skill feedback and planning patterns, reads closed bead outcomes, and spawns an analysis agent per skill. Each agent writes proposed changes to a docs file for your review before anything is applied.

How does the skill coordinate multiple analysis agents?▼

It bootstraps Agent Mail with macro_start_session, creates a refine-skills team, and spawns one background general-purpose agent per skill. Agents report results via send_message, and the coordinator shuts each down individually after completion.

Does it modify SKILL.md files automatically?▼

No. Proposed changes are written to docs/skill-refine-<name>-proposed.md files first. The coordinator presents findings per skill and asks which skills to update, applying edits only to approved SKILL.md files.

What data sources does the skill analysis use?▼

It uses agent-mail message history searched with queries like skill feedback and planning pattern, plus closed bead data from br list --json including review feedback. These ground the proposed improvements in actual session outcomes.

What happens if an analysis agent becomes idle or stuck?▼

The coordinator saves each task ID for potential TaskStop use and nudges idle agents individually by name. Shutdown messages are sent per agent, never broadcast to all agents at once.

Related Skills

Explore More Software Engineering Skills→
Dunc4nJDunc4nJ

vibing-with-ntm

Coordinate multi-agent swarms across NTM, Agent Mail, Beads, and BV.

Community
Advanced
darellchua2darellchua2
4

opencode-skills-maintainer

Scan skills/ for SKILL.md files and update agent prompts.

Community
Advanced
vnnklvnnkl
2

bead-polishing-ralph

Polish Ralph-TUI beads to fit one agent context window.

Community
Advanced
liminal-ailiminal-ai

full-code-review

Orchestrate six parallel specialized code reviewers into a single verified report.

Official
Advanced