Skill

improve-expertise

From tac

Syncs domain expert's expertise.yaml with codebase via self-improve agent, git diff checks. Invoke /improve-expertise <domain> after changes or periodically.

Bash

Git

automation

developer-tools

npx claudepluginhub melodic-software/claude-code-plugins --plugin tac

Tool Access

This skill is limited to using the following tools:

ReadWriteEditGlobGrepBash

Preview

Run the self-improve workflow on an agent expert's expertise file to maintain accuracy.

SKILL.md

Similar Skills

self-improve-prompt-design

Designs self-improve prompts to sync agent expertise files with codebase via git diff checks, validation, and auto-updates. For Act-Learn-Reuse LEARN step and expert maintenance.

4 tools

tac

upgrade

3.3k

Proposes upgrades to generated skills via Ars Contexta research graph consultation, justifying methodology improvements. Never auto-implements; triggers on /upgrade or upgrade phrases.

1 file6 tools

arscontexta

distill

Distills patterns from Claude Code work history, git logs, lessons, and memory to suggest new agents/skills, review roster quality, prune redundancies, or consolidate feedback into rules.

8 tools

foundry

Stats

Parent Repo Stars63

Parent Repo Forks9

Last CommitFeb 15, 2026

Actions

View Source View Plugin View on GitHub View README

Help us improve

Share bugs, ideas, or general feedback.

Improve Expertise

Run the self-improve workflow on an agent expert's expertise file to maintain accuracy.

Arguments

$1: Domain name (required, e.g., "database", "websocket")
$2: Check git diff flag (optional, "true" or "false", default: "true")

Instructions

You are running the self-improve workflow to sync an expert's mental model with the actual codebase.

Step 1: Parse Arguments

Extract:

Domain name from $1 (required)
Check git diff flag from $2 (optional, default: true)

If no domain provided, STOP and ask for domain name.

Step 2: Validate Expert Exists

Check if expert directory exists using Glob:

Glob: .claude/commands/experts/{$1}/expertise.yaml

If not found:

STOP and report "Expert not found. Use /tac:create-expert to create it first."
List available experts using: Glob: .claude/commands/experts/*

Step 3: Spawn Self-Improver Agent

Delegate to the self-improver agent with:

Domain name
Path to expertise file: .claude/commands/experts/{domain}/expertise.yaml
Check git diff flag

The agent will:

Check git diff (if flag is true)
Read current expertise
Validate against codebase
Identify discrepancies
Update expertise file
Enforce line limits
Validate output

Step 4: Report Results

Display the self-improve report:

## Self-Improve Complete: {domain}

### Changes Made

- [List of changes]

### Expertise Health

| Metric | Value |
| --- | --- |
| Line count | X/1000 |
| Files validated | X/X exist |
| Functions verified | X/X accurate |

### Recommendations

- [Any recommendations for manual review]

Quick Usage

# Sync after making changes (checks git diff)
/tac:improve-expertise database true

# Full rescan without git diff check
/tac:improve-expertise database false

# Default behavior (checks git diff)
/tac:improve-expertise websocket

When to Run

Trigger	Command
After any build/fix work	`/tac:improve-expertise {domain} true`
Periodic maintenance	`/tac:improve-expertise {domain} false`
Suspect drift	`/tac:improve-expertise {domain} false`
Before major planning	`/tac:improve-expertise {domain} false`

Notes

This is the LEARN step of Act-Learn-Reuse
Run after every ACT (build, fix, modify) step
Mental model is NOT source of truth - this syncs it with code
Expertise file must stay under 1000 lines

Last Updated: 2025-12-15