improving-jarvis

erikpr1994's avatarfrom erikpr1994

Use when identifying opportunities to improve the Jarvis system, adding patterns, creating skills, or updating rules based on repeated guidance.

⭐0stars🔀0forks📁View on GitHub🕐Updated Jan 9, 2026
[Prompt Templates and Packaging]

When & Why to Use This Skill

The 'Improving Jarvis' skill is a meta-utility designed to facilitate the continuous evolution and optimization of the Jarvis AI agent. By monitoring for repeated instructions and workflow inefficiencies, it enables the seamless creation of new skills, the documentation of reusable patterns, and the refinement of system rules. This ensures the agent remains aligned with user needs, maintains high consistency, and adheres to project-specific standards through a structured validation and rollback framework, effectively solving the problem of manual system maintenance and agent drift.

Use Cases

  • Capturing Recurring Logic: Automatically transform manual instructions given 3+ times into permanent system rules to eliminate repetitive prompting and ensure long-term consistency.
  • Workflow Formalization: Convert complex, multi-step manual processes into structured 'Skills' using a TDD (Test-Driven Development) approach to ensure high reliability and performance.
  • Project Convention Alignment: Dynamically update CLAUDE.md and internal style guides as project requirements evolve, keeping the agent's behavior perfectly synced with the latest coding standards.
  • Quality Assurance for Agent Logic: Implement a rigorous validation process for new agent capabilities, monitoring performance for 3 sessions and maintaining a false-positive rate below 10%.
nameimproving-jarvis
description"Use when identifying opportunities to improve the Jarvis system, adding patterns, creating skills, or updating rules based on repeated guidance."

Improving Jarvis

Overview

Jarvis improves through captured patterns, new skills, and refined rules. This skill guides when and how to enhance the system.

When to Trigger

Automatic triggers:

  • Same instruction given manually 3+ times
  • Discovered pattern not in library
  • Workflow inefficiency noticed
  • User explicitly requests improvement

Ask user: "This [pattern/workaround] seems reusable. Should I capture it?"

Improvement Types

1. Add Pattern

When: Reusable solution discovered Action:

  1. Document in patterns/ with keywords
  2. Add to pattern index
  3. Test with 3 example prompts

2. Create Skill

When: Complex workflow needs guidance Action:

  1. Follow writing-skills (TDD approach)
  2. Run RED-GREEN-REFACTOR cycle
  3. Add to skill-rules.json

3. Update Rule

When: Behavior needs consistency Action:

  1. Modify existing rule or create new
  2. Test for false positive rate < 10%
  3. Preserve backward compatibility

4. Add Hook

When: Automation should apply to all events Action:

  1. Follow writing-hooks guide
  2. Test edge cases and timeouts
  3. Add graceful failure handling

5. Update CLAUDE.md

When: Project conventions change Action:

  1. Follow writing-claude-md guide
  2. Respect token budgets
  3. Test with real prompts

Validation Process

Create component following guide
        |
        v
Run component-specific tests
        |
        v
Monitor for 3 sessions
        |
        v
Issues found? --> Fix and retest
        |
        v (no issues)
Mark as stable

Rollback Triggers

Auto-rollback if:

  • False positive rate > 10%
  • User reverts change manually
  • Component causes quality degradation
  • Test failures after change

Rollback process:

  1. Revert to previous version
  2. Document what went wrong
  3. Analyze root cause
  4. Create improved version with lessons

Quality Checklist

Before marking improvement complete:

  • Followed relevant writing-* skill
  • Tested with real scenarios
  • No regression in existing functionality
  • Token budget respected
  • Documented in appropriate location
  • skill-rules.json updated (if skill)
  • User approved improvement

Improvement Tracking

Log improvements in .claude/improvements.log:

YYYY-MM-DD | TYPE | NAME | REASON
2025-01-04 | skill | git-expert | Repeated git guidance
2025-01-04 | pattern | optimistic-updates | Discovered in feature work

Common Mistakes

Mistake Fix
Improve without testing Run validation process
Skip user approval Always confirm significant changes
Forget skill-rules.json Update triggers for new skills
Too aggressive rollout Monitor for 3 sessions first
No documentation Log all improvements
Ignore false positives Track and maintain < 10% rate