skills.rest
ToanPV90
ToanPV90COMMUNITY
Explore all skills by this author →
GitHub

improve-agent

Improves Claude Code agents by analyzing feedback and proposing validated prompt edits.

Updated Apr 15, 2026
Search Tags:#version control#prompt engineering#Claude Code#agent memory#feedback analysis#agent tuning#routing logs
One-click install
npx skills add https://github.com/ToanPV90/dotfiles --skill improve-agent-toanpv90
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: improve-agent
Source: https://github.com/ToanPV90/dotfiles/tree/main/claude/.claude/skills/improve-agent
Command: npx skills add https://github.com/ToanPV90/dotfiles --skill improve-agent-toanpv90

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It helps you improve Claude Code agent performance by turning accumulated feedback and routing outcomes into concrete, targeted edits to the agent definition instead of guesswork.

Core Features & Use Cases

  • Performance analysis from feedback: Reviews the agent definition, feedback-memory patterns, and routing logs to identify the most frequent failure modes (e.g., wrong tool choice, incomplete output, misunderstanding intent, hallucinations, excessive scope).
  • Surgical prompt engineering: Proposes minimal, traceable changes mapped to specific failure modes, including constraints and verification/checklists to prevent regressions.
  • Validation and safe rollout: Runs representative tasks to confirm measurable improvement, restores from backup on failure, and records the change via a committed update and a structured memory entry.

Quick Start

Use the improve-agent skill when you tell the system “Improve agent improve-agent” after you notice repeated corrections or provide performance feedback for that agent.

Frequently Asked Questions about improve-agent

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

FAQPage Schema
How do I improve Claude Code agent performance using accumulated feedback?▼

Improve Claude Code agent performance by analyzing existing feedback patterns and routing logs to identify failure modes, then proposing targeted prompt edits validated against representative tasks.

What is the process for tuning agent prompts to fix repeated correction patterns?▼

Tuning agent prompts involves a four-phase workflow: baseline measurement, surgical prompt diffs, representative task validation, and version-controlled updates with an improvement record in agent memory.

When should I schedule an agent prompt review for task-router accuracy?▼

Schedule an agent prompt review when you notice declining task-router accuracy, repeated correction patterns, or as a routine monthly check to maintain performance.

How do I prevent regressions when applying surgical prompt engineering to agent definitions?▼

Prevent regressions during surgical prompt engineering by running representative tasks to confirm measurable improvement, adding constraints and verification checklists, and restoring from backup on failure.

Can I apply version control to Claude Code agent memory updates?▼

Yes, apply version control to agent memory updates by committing the changes alongside a structured memory entry that records the improvement, ensuring traceability.

What failure modes can I identify from agent routing logs and feedback analysis?▼

Agent routing logs and feedback analysis identify frequent failure modes including wrong tool choice, incomplete output, misunderstanding intent, hallucinations, and excessive scope.

Related Skills

Explore More Software Engineering Skills→
DhumitechDhumitech

agent-evaluation

Automate AI agent evaluation and benchmarking with repeatable test pipelines.

Community
Advanced
mkalhitti-cloudmkalhitti-cloud
2

context-transfer

Generate structured session handoff prompts for cross-environment context preservation.

Community
Advanced
SergeiGolosSergeiGolos
1

performance-profiling

Audit web performance with Lighthouse and analyze bundle sizes.

Community
Intermediate
imonmiimonmi

benchmark

Compare baseline and PR performance metrics to detect regressions.

Community
Advanced